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.
135 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
136 sv, av, hv...) contains type and reference count information, and for
137 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
138 contains fields specific to each type. Some types store all they need
139 in the head, so don't have a body.
141 In all but the most memory-paranoid configurations (ex: PURIFY), heads
142 and bodies are allocated out of arenas, which by default are
143 approximately 4K chunks of memory parcelled up into N heads or bodies.
144 Sv-bodies are allocated by their sv-type, guaranteeing size
145 consistency needed to allocate safely from arrays.
147 For SV-heads, the first slot in each arena is reserved, and holds a
148 link to the next arena, some flags, and a note of the number of slots.
149 Snaked through each arena chain is a linked list of free items; when
150 this becomes empty, an extra arena is allocated and divided up into N
151 items which are threaded into the free list.
153 SV-bodies are similar, but they use arena-sets by default, which
154 separate the link and info from the arena itself, and reclaim the 1st
155 slot in the arena. SV-bodies are further described later.
157 The following global variables are associated with arenas:
159 PL_sv_arenaroot pointer to list of SV arenas
160 PL_sv_root pointer to list of free SV structures
162 PL_body_arenas head of linked-list of body arenas
163 PL_body_roots[] array of pointers to list of free bodies of svtype
164 arrays are indexed by the svtype needed
166 A few special SV heads are not allocated from an arena, but are
167 instead directly created in the interpreter structure, eg PL_sv_undef.
168 The size of arenas can be changed from the default by setting
169 PERL_ARENA_SIZE appropriately at compile time.
171 The SV arena serves the secondary purpose of allowing still-live SVs
172 to be located and destroyed during final cleanup.
174 At the lowest level, the macros new_SV() and del_SV() grab and free
175 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
176 to return the SV to the free list with error checking.) new_SV() calls
177 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
178 SVs in the free list have their SvTYPE field set to all ones.
180 At the time of very final cleanup, sv_free_arenas() is called from
181 perl_destruct() to physically free all the arenas allocated since the
182 start of the interpreter.
184 The function visit() scans the SV arenas list, and calls a specified
185 function for each SV it finds which is still live - ie which has an SvTYPE
186 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
187 following functions (specified as [function that calls visit()] / [function
188 called by visit() for each SV]):
190 sv_report_used() / do_report_used()
191 dump all remaining SVs (debugging aid)
193 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
194 do_clean_named_io_objs(),do_curse()
195 Attempt to free all objects pointed to by RVs,
196 try to do the same for all objects indir-
197 ectly referenced by typeglobs too, and
198 then do a final sweep, cursing any
199 objects that remain. Called once from
200 perl_destruct(), prior to calling sv_clean_all()
203 sv_clean_all() / do_clean_all()
204 SvREFCNT_dec(sv) each remaining SV, possibly
205 triggering an sv_free(). It also sets the
206 SVf_BREAK flag on the SV to indicate that the
207 refcnt has been artificially lowered, and thus
208 stopping sv_free() from giving spurious warnings
209 about SVs which unexpectedly have a refcnt
210 of zero. called repeatedly from perl_destruct()
211 until there are no SVs left.
213 =head2 Arena allocator API Summary
215 Private API to rest of sv.c
219 new_XPVNV(), del_XPVGV(),
224 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
228 * ========================================================================= */
231 * "A time to plant, and a time to uproot what was planted..."
235 # define MEM_LOG_NEW_SV(sv, file, line, func) \
236 Perl_mem_log_new_sv(sv, file, line, func)
237 # define MEM_LOG_DEL_SV(sv, file, line, func) \
238 Perl_mem_log_del_sv(sv, file, line, func)
240 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
241 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
244 #ifdef DEBUG_LEAKING_SCALARS
245 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
246 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
248 # define DEBUG_SV_SERIAL(sv) \
249 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
250 PTR2UV(sv), (long)(sv)->sv_debug_serial))
252 # define FREE_SV_DEBUG_FILE(sv)
253 # define DEBUG_SV_SERIAL(sv) NOOP
257 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
258 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
259 /* Whilst I'd love to do this, it seems that things like to check on
261 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
263 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
264 PoisonNew(&SvREFCNT(sv), 1, U32)
266 # define SvARENA_CHAIN(sv) SvANY(sv)
267 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
268 # define POISON_SV_HEAD(sv)
271 /* Mark an SV head as unused, and add to free list.
273 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
274 * its refcount artificially decremented during global destruction, so
275 * there may be dangling pointers to it. The last thing we want in that
276 * case is for it to be reused. */
278 #define plant_SV(p) \
280 const U32 old_flags = SvFLAGS(p); \
281 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
282 DEBUG_SV_SERIAL(p); \
283 FREE_SV_DEBUG_FILE(p); \
285 SvFLAGS(p) = SVTYPEMASK; \
286 if (!(old_flags & SVf_BREAK)) { \
287 SvARENA_CHAIN_SET(p, PL_sv_root); \
293 #define uproot_SV(p) \
296 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
301 /* make some more SVs by adding another arena */
307 char *chunk; /* must use New here to match call to */
308 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
309 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
314 /* new_SV(): return a new, empty SV head */
316 #ifdef DEBUG_LEAKING_SCALARS
317 /* provide a real function for a debugger to play with */
319 S_new_SV(pTHX_ const char *file, int line, const char *func)
326 sv = S_more_sv(aTHX);
330 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
331 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
337 sv->sv_debug_inpad = 0;
338 sv->sv_debug_parent = NULL;
339 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
341 sv->sv_debug_serial = PL_sv_serial++;
343 MEM_LOG_NEW_SV(sv, file, line, func);
344 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
345 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
349 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
357 (p) = S_more_sv(aTHX); \
361 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
366 /* del_SV(): return an empty SV head to the free list */
379 S_del_sv(pTHX_ SV *p)
381 PERL_ARGS_ASSERT_DEL_SV;
386 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
387 const SV * const sv = sva + 1;
388 const SV * const svend = &sva[SvREFCNT(sva)];
389 if (p >= sv && p < svend) {
395 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
396 "Attempt to free non-arena SV: 0x%" UVxf
397 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
404 #else /* ! DEBUGGING */
406 #define del_SV(p) plant_SV(p)
408 #endif /* DEBUGGING */
412 =head1 SV Manipulation Functions
414 =for apidoc sv_add_arena
416 Given a chunk of memory, link it to the head of the list of arenas,
417 and split it into a list of free SVs.
423 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
425 SV *const sva = MUTABLE_SV(ptr);
429 PERL_ARGS_ASSERT_SV_ADD_ARENA;
431 /* The first SV in an arena isn't an SV. */
432 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
433 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
434 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
436 PL_sv_arenaroot = sva;
437 PL_sv_root = sva + 1;
439 svend = &sva[SvREFCNT(sva) - 1];
442 SvARENA_CHAIN_SET(sv, (sv + 1));
446 /* Must always set typemask because it's always checked in on cleanup
447 when the arenas are walked looking for objects. */
448 SvFLAGS(sv) = SVTYPEMASK;
451 SvARENA_CHAIN_SET(sv, 0);
455 SvFLAGS(sv) = SVTYPEMASK;
458 /* visit(): call the named function for each non-free SV in the arenas
459 * whose flags field matches the flags/mask args. */
462 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
467 PERL_ARGS_ASSERT_VISIT;
469 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
470 const SV * const svend = &sva[SvREFCNT(sva)];
472 for (sv = sva + 1; sv < svend; ++sv) {
473 if (SvTYPE(sv) != (svtype)SVTYPEMASK
474 && (sv->sv_flags & mask) == flags
487 /* called by sv_report_used() for each live SV */
490 do_report_used(pTHX_ SV *const sv)
492 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
493 PerlIO_printf(Perl_debug_log, "****\n");
500 =for apidoc sv_report_used
502 Dump the contents of all SVs not yet freed (debugging aid).
508 Perl_sv_report_used(pTHX)
511 visit(do_report_used, 0, 0);
517 /* called by sv_clean_objs() for each live SV */
520 do_clean_objs(pTHX_ SV *const ref)
524 SV * const target = SvRV(ref);
525 if (SvOBJECT(target)) {
526 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
527 if (SvWEAKREF(ref)) {
528 sv_del_backref(target, ref);
534 SvREFCNT_dec_NN(target);
541 /* clear any slots in a GV which hold objects - except IO;
542 * called by sv_clean_objs() for each live GV */
545 do_clean_named_objs(pTHX_ SV *const sv)
548 assert(SvTYPE(sv) == SVt_PVGV);
549 assert(isGV_with_GP(sv));
553 /* freeing GP entries may indirectly free the current GV;
554 * hold onto it while we mess with the GP slots */
557 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
558 DEBUG_D((PerlIO_printf(Perl_debug_log,
559 "Cleaning named glob SV object:\n "), sv_dump(obj)));
561 SvREFCNT_dec_NN(obj);
563 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
564 DEBUG_D((PerlIO_printf(Perl_debug_log,
565 "Cleaning named glob AV object:\n "), sv_dump(obj)));
567 SvREFCNT_dec_NN(obj);
569 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
570 DEBUG_D((PerlIO_printf(Perl_debug_log,
571 "Cleaning named glob HV object:\n "), sv_dump(obj)));
573 SvREFCNT_dec_NN(obj);
575 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
576 DEBUG_D((PerlIO_printf(Perl_debug_log,
577 "Cleaning named glob CV object:\n "), sv_dump(obj)));
579 SvREFCNT_dec_NN(obj);
581 SvREFCNT_dec_NN(sv); /* undo the inc above */
584 /* clear any IO slots in a GV which hold objects (except stderr, defout);
585 * called by sv_clean_objs() for each live GV */
588 do_clean_named_io_objs(pTHX_ SV *const sv)
591 assert(SvTYPE(sv) == SVt_PVGV);
592 assert(isGV_with_GP(sv));
593 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
597 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
598 DEBUG_D((PerlIO_printf(Perl_debug_log,
599 "Cleaning named glob IO object:\n "), sv_dump(obj)));
601 SvREFCNT_dec_NN(obj);
603 SvREFCNT_dec_NN(sv); /* undo the inc above */
606 /* Void wrapper to pass to visit() */
608 do_curse(pTHX_ SV * const sv) {
609 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
610 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
616 =for apidoc sv_clean_objs
618 Attempt to destroy all objects not yet freed.
624 Perl_sv_clean_objs(pTHX)
627 PL_in_clean_objs = TRUE;
628 visit(do_clean_objs, SVf_ROK, SVf_ROK);
629 /* Some barnacles may yet remain, clinging to typeglobs.
630 * Run the non-IO destructors first: they may want to output
631 * error messages, close files etc */
632 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
634 /* And if there are some very tenacious barnacles clinging to arrays,
635 closures, or what have you.... */
636 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
637 olddef = PL_defoutgv;
638 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
639 if (olddef && isGV_with_GP(olddef))
640 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
641 olderr = PL_stderrgv;
642 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
643 if (olderr && isGV_with_GP(olderr))
644 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
645 SvREFCNT_dec(olddef);
646 PL_in_clean_objs = FALSE;
649 /* called by sv_clean_all() for each live SV */
652 do_clean_all(pTHX_ SV *const sv)
654 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
655 /* don't clean pid table and strtab */
658 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
659 SvFLAGS(sv) |= SVf_BREAK;
664 =for apidoc sv_clean_all
666 Decrement the refcnt of each remaining SV, possibly triggering a
667 cleanup. This function may have to be called multiple times to free
668 SVs which are in complex self-referential hierarchies.
674 Perl_sv_clean_all(pTHX)
677 PL_in_clean_all = TRUE;
678 cleaned = visit(do_clean_all, 0,0);
683 ARENASETS: a meta-arena implementation which separates arena-info
684 into struct arena_set, which contains an array of struct
685 arena_descs, each holding info for a single arena. By separating
686 the meta-info from the arena, we recover the 1st slot, formerly
687 borrowed for list management. The arena_set is about the size of an
688 arena, avoiding the needless malloc overhead of a naive linked-list.
690 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
691 memory in the last arena-set (1/2 on average). In trade, we get
692 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
693 smaller types). The recovery of the wasted space allows use of
694 small arenas for large, rare body types, by changing array* fields
695 in body_details_by_type[] below.
698 char *arena; /* the raw storage, allocated aligned */
699 size_t size; /* its size ~4k typ */
700 svtype utype; /* bodytype stored in arena */
705 /* Get the maximum number of elements in set[] such that struct arena_set
706 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
707 therefore likely to be 1 aligned memory page. */
709 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
710 - 2 * sizeof(int)) / sizeof (struct arena_desc))
713 struct arena_set* next;
714 unsigned int set_size; /* ie ARENAS_PER_SET */
715 unsigned int curr; /* index of next available arena-desc */
716 struct arena_desc set[ARENAS_PER_SET];
720 =for apidoc sv_free_arenas
722 Deallocate the memory used by all arenas. Note that all the individual SV
723 heads and bodies within the arenas must already have been freed.
729 Perl_sv_free_arenas(pTHX)
735 /* Free arenas here, but be careful about fake ones. (We assume
736 contiguity of the fake ones with the corresponding real ones.) */
738 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
739 svanext = MUTABLE_SV(SvANY(sva));
740 while (svanext && SvFAKE(svanext))
741 svanext = MUTABLE_SV(SvANY(svanext));
748 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
751 struct arena_set *current = aroot;
754 assert(aroot->set[i].arena);
755 Safefree(aroot->set[i].arena);
763 i = PERL_ARENA_ROOTS_SIZE;
765 PL_body_roots[i] = 0;
772 Here are mid-level routines that manage the allocation of bodies out
773 of the various arenas. There are 4 kinds of arenas:
775 1. SV-head arenas, which are discussed and handled above
776 2. regular body arenas
777 3. arenas for reduced-size bodies
780 Arena types 2 & 3 are chained by body-type off an array of
781 arena-root pointers, which is indexed by svtype. Some of the
782 larger/less used body types are malloced singly, since a large
783 unused block of them is wasteful. Also, several svtypes dont have
784 bodies; the data fits into the sv-head itself. The arena-root
785 pointer thus has a few unused root-pointers (which may be hijacked
786 later for arena type 4)
788 3 differs from 2 as an optimization; some body types have several
789 unused fields in the front of the structure (which are kept in-place
790 for consistency). These bodies can be allocated in smaller chunks,
791 because the leading fields arent accessed. Pointers to such bodies
792 are decremented to point at the unused 'ghost' memory, knowing that
793 the pointers are used with offsets to the real memory.
795 Allocation of SV-bodies is similar to SV-heads, differing as follows;
796 the allocation mechanism is used for many body types, so is somewhat
797 more complicated, it uses arena-sets, and has no need for still-live
800 At the outermost level, (new|del)_X*V macros return bodies of the
801 appropriate type. These macros call either (new|del)_body_type or
802 (new|del)_body_allocated macro pairs, depending on specifics of the
803 type. Most body types use the former pair, the latter pair is used to
804 allocate body types with "ghost fields".
806 "ghost fields" are fields that are unused in certain types, and
807 consequently don't need to actually exist. They are declared because
808 they're part of a "base type", which allows use of functions as
809 methods. The simplest examples are AVs and HVs, 2 aggregate types
810 which don't use the fields which support SCALAR semantics.
812 For these types, the arenas are carved up into appropriately sized
813 chunks, we thus avoid wasted memory for those unaccessed members.
814 When bodies are allocated, we adjust the pointer back in memory by the
815 size of the part not allocated, so it's as if we allocated the full
816 structure. (But things will all go boom if you write to the part that
817 is "not there", because you'll be overwriting the last members of the
818 preceding structure in memory.)
820 We calculate the correction using the STRUCT_OFFSET macro on the first
821 member present. If the allocated structure is smaller (no initial NV
822 actually allocated) then the net effect is to subtract the size of the NV
823 from the pointer, to return a new pointer as if an initial NV were actually
824 allocated. (We were using structures named *_allocated for this, but
825 this turned out to be a subtle bug, because a structure without an NV
826 could have a lower alignment constraint, but the compiler is allowed to
827 optimised accesses based on the alignment constraint of the actual pointer
828 to the full structure, for example, using a single 64 bit load instruction
829 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
831 This is the same trick as was used for NV and IV bodies. Ironically it
832 doesn't need to be used for NV bodies any more, because NV is now at
833 the start of the structure. IV bodies, and also in some builds NV bodies,
834 don't need it either, because they are no longer allocated.
836 In turn, the new_body_* allocators call S_new_body(), which invokes
837 new_body_inline macro, which takes a lock, and takes a body off the
838 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
839 necessary to refresh an empty list. Then the lock is released, and
840 the body is returned.
842 Perl_more_bodies allocates a new arena, and carves it up into an array of N
843 bodies, which it strings into a linked list. It looks up arena-size
844 and body-size from the body_details table described below, thus
845 supporting the multiple body-types.
847 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
848 the (new|del)_X*V macros are mapped directly to malloc/free.
850 For each sv-type, struct body_details bodies_by_type[] carries
851 parameters which control these aspects of SV handling:
853 Arena_size determines whether arenas are used for this body type, and if
854 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
855 zero, forcing individual mallocs and frees.
857 Body_size determines how big a body is, and therefore how many fit into
858 each arena. Offset carries the body-pointer adjustment needed for
859 "ghost fields", and is used in *_allocated macros.
861 But its main purpose is to parameterize info needed in
862 Perl_sv_upgrade(). The info here dramatically simplifies the function
863 vs the implementation in 5.8.8, making it table-driven. All fields
864 are used for this, except for arena_size.
866 For the sv-types that have no bodies, arenas are not used, so those
867 PL_body_roots[sv_type] are unused, and can be overloaded. In
868 something of a special case, SVt_NULL is borrowed for HE arenas;
869 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
870 bodies_by_type[SVt_NULL] slot is not used, as the table is not
875 struct body_details {
876 U8 body_size; /* Size to allocate */
877 U8 copy; /* Size of structure to copy (may be shorter) */
878 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
879 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
880 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
881 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
882 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
883 U32 arena_size; /* Size of arena to allocate */
891 /* With -DPURFIY we allocate everything directly, and don't use arenas.
892 This seems a rather elegant way to simplify some of the code below. */
893 #define HASARENA FALSE
895 #define HASARENA TRUE
897 #define NOARENA FALSE
899 /* Size the arenas to exactly fit a given number of bodies. A count
900 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
901 simplifying the default. If count > 0, the arena is sized to fit
902 only that many bodies, allowing arenas to be used for large, rare
903 bodies (XPVFM, XPVIO) without undue waste. The arena size is
904 limited by PERL_ARENA_SIZE, so we can safely oversize the
907 #define FIT_ARENA0(body_size) \
908 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
909 #define FIT_ARENAn(count,body_size) \
910 ( count * body_size <= PERL_ARENA_SIZE) \
911 ? count * body_size \
912 : FIT_ARENA0 (body_size)
913 #define FIT_ARENA(count,body_size) \
915 ? FIT_ARENAn (count, body_size) \
916 : FIT_ARENA0 (body_size))
918 /* Calculate the length to copy. Specifically work out the length less any
919 final padding the compiler needed to add. See the comment in sv_upgrade
920 for why copying the padding proved to be a bug. */
922 #define copy_length(type, last_member) \
923 STRUCT_OFFSET(type, last_member) \
924 + sizeof (((type*)SvANY((const SV *)0))->last_member)
926 static const struct body_details bodies_by_type[] = {
927 /* HEs use this offset for their arena. */
928 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
930 /* IVs are in the head, so the allocation size is 0. */
932 sizeof(IV), /* This is used to copy out the IV body. */
933 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
934 NOARENA /* IVS don't need an arena */, 0
939 STRUCT_OFFSET(XPVNV, xnv_u),
940 SVt_NV, FALSE, HADNV, NOARENA, 0 },
942 { sizeof(NV), sizeof(NV),
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
947 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
948 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_PV, FALSE, NONV, HASARENA,
951 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
953 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_INVLIST, TRUE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
959 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVIV, FALSE, NONV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
965 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
966 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
967 + STRUCT_OFFSET(XPV, xpv_cur),
968 SVt_PVNV, FALSE, HADNV, HASARENA,
969 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
971 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
972 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
977 SVt_REGEXP, TRUE, NONV, HASARENA,
978 FIT_ARENA(0, sizeof(regexp))
981 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
982 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
984 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
985 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
988 copy_length(XPVAV, xav_alloc),
990 SVt_PVAV, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XPVAV)) },
994 copy_length(XPVHV, xhv_max),
996 SVt_PVHV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVHV)) },
1002 SVt_PVCV, TRUE, NONV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVCV)) },
1008 SVt_PVFM, TRUE, NONV, NOARENA,
1009 FIT_ARENA(20, sizeof(XPVFM)) },
1014 SVt_PVIO, TRUE, NONV, HASARENA,
1015 FIT_ARENA(24, sizeof(XPVIO)) },
1018 #define new_body_allocated(sv_type) \
1019 (void *)((char *)S_new_body(aTHX_ sv_type) \
1020 - bodies_by_type[sv_type].offset)
1022 /* return a thing to the free list */
1024 #define del_body(thing, root) \
1026 void ** const thing_copy = (void **)thing; \
1027 *thing_copy = *root; \
1028 *root = (void*)thing_copy; \
1032 #if !(NVSIZE <= IVSIZE)
1033 # define new_XNV() safemalloc(sizeof(XPVNV))
1035 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1036 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1038 #define del_XPVGV(p) safefree(p)
1042 #if !(NVSIZE <= IVSIZE)
1043 # define new_XNV() new_body_allocated(SVt_NV)
1045 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1046 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1048 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1049 &PL_body_roots[SVt_PVGV])
1053 /* no arena for you! */
1055 #define new_NOARENA(details) \
1056 safemalloc((details)->body_size + (details)->offset)
1057 #define new_NOARENAZ(details) \
1058 safecalloc((details)->body_size + (details)->offset, 1)
1061 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1062 const size_t arena_size)
1064 void ** const root = &PL_body_roots[sv_type];
1065 struct arena_desc *adesc;
1066 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1070 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1071 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1074 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT)
1075 static bool done_sanity_check;
1077 /* PERL_GLOBAL_STRUCT cannot coexist with global
1078 * variables like done_sanity_check. */
1079 if (!done_sanity_check) {
1080 unsigned int i = SVt_LAST;
1082 done_sanity_check = TRUE;
1085 assert (bodies_by_type[i].type == i);
1091 /* may need new arena-set to hold new arena */
1092 if (!aroot || aroot->curr >= aroot->set_size) {
1093 struct arena_set *newroot;
1094 Newxz(newroot, 1, struct arena_set);
1095 newroot->set_size = ARENAS_PER_SET;
1096 newroot->next = aroot;
1098 PL_body_arenas = (void *) newroot;
1099 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1102 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1103 curr = aroot->curr++;
1104 adesc = &(aroot->set[curr]);
1105 assert(!adesc->arena);
1107 Newx(adesc->arena, good_arena_size, char);
1108 adesc->size = good_arena_size;
1109 adesc->utype = sv_type;
1110 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1111 curr, (void*)adesc->arena, (UV)good_arena_size));
1113 start = (char *) adesc->arena;
1115 /* Get the address of the byte after the end of the last body we can fit.
1116 Remember, this is integer division: */
1117 end = start + good_arena_size / body_size * body_size;
1119 /* computed count doesn't reflect the 1st slot reservation */
1120 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1121 DEBUG_m(PerlIO_printf(Perl_debug_log,
1122 "arena %p end %p arena-size %d (from %d) type %d "
1124 (void*)start, (void*)end, (int)good_arena_size,
1125 (int)arena_size, sv_type, (int)body_size,
1126 (int)good_arena_size / (int)body_size));
1128 DEBUG_m(PerlIO_printf(Perl_debug_log,
1129 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1130 (void*)start, (void*)end,
1131 (int)arena_size, sv_type, (int)body_size,
1132 (int)good_arena_size / (int)body_size));
1134 *root = (void *)start;
1137 /* Where the next body would start: */
1138 char * const next = start + body_size;
1141 /* This is the last body: */
1142 assert(next == end);
1144 *(void **)start = 0;
1148 *(void**) start = (void *)next;
1153 /* grab a new thing from the free list, allocating more if necessary.
1154 The inline version is used for speed in hot routines, and the
1155 function using it serves the rest (unless PURIFY).
1157 #define new_body_inline(xpv, sv_type) \
1159 void ** const r3wt = &PL_body_roots[sv_type]; \
1160 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1161 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1162 bodies_by_type[sv_type].body_size,\
1163 bodies_by_type[sv_type].arena_size)); \
1164 *(r3wt) = *(void**)(xpv); \
1170 S_new_body(pTHX_ const svtype sv_type)
1173 new_body_inline(xpv, sv_type);
1179 static const struct body_details fake_rv =
1180 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1183 =for apidoc sv_upgrade
1185 Upgrade an SV to a more complex form. Generally adds a new body type to the
1186 SV, then copies across as much information as possible from the old body.
1187 It croaks if the SV is already in a more complex form than requested. You
1188 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1189 before calling C<sv_upgrade>, and hence does not croak. See also
1196 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1200 const svtype old_type = SvTYPE(sv);
1201 const struct body_details *new_type_details;
1202 const struct body_details *old_type_details
1203 = bodies_by_type + old_type;
1204 SV *referent = NULL;
1206 PERL_ARGS_ASSERT_SV_UPGRADE;
1208 if (old_type == new_type)
1211 /* This clause was purposefully added ahead of the early return above to
1212 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1213 inference by Nick I-S that it would fix other troublesome cases. See
1214 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1216 Given that shared hash key scalars are no longer PVIV, but PV, there is
1217 no longer need to unshare so as to free up the IVX slot for its proper
1218 purpose. So it's safe to move the early return earlier. */
1220 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1221 sv_force_normal_flags(sv, 0);
1224 old_body = SvANY(sv);
1226 /* Copying structures onto other structures that have been neatly zeroed
1227 has a subtle gotcha. Consider XPVMG
1229 +------+------+------+------+------+-------+-------+
1230 | NV | CUR | LEN | IV | MAGIC | STASH |
1231 +------+------+------+------+------+-------+-------+
1232 0 4 8 12 16 20 24 28
1234 where NVs are aligned to 8 bytes, so that sizeof that structure is
1235 actually 32 bytes long, with 4 bytes of padding at the end:
1237 +------+------+------+------+------+-------+-------+------+
1238 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1239 +------+------+------+------+------+-------+-------+------+
1240 0 4 8 12 16 20 24 28 32
1242 so what happens if you allocate memory for this structure:
1244 +------+------+------+------+------+-------+-------+------+------+...
1245 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1246 +------+------+------+------+------+-------+-------+------+------+...
1247 0 4 8 12 16 20 24 28 32 36
1249 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1250 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1251 started out as zero once, but it's quite possible that it isn't. So now,
1252 rather than a nicely zeroed GP, you have it pointing somewhere random.
1255 (In fact, GP ends up pointing at a previous GP structure, because the
1256 principle cause of the padding in XPVMG getting garbage is a copy of
1257 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1258 this happens to be moot because XPVGV has been re-ordered, with GP
1259 no longer after STASH)
1261 So we are careful and work out the size of used parts of all the
1269 referent = SvRV(sv);
1270 old_type_details = &fake_rv;
1271 if (new_type == SVt_NV)
1272 new_type = SVt_PVNV;
1274 if (new_type < SVt_PVIV) {
1275 new_type = (new_type == SVt_NV)
1276 ? SVt_PVNV : SVt_PVIV;
1281 if (new_type < SVt_PVNV) {
1282 new_type = SVt_PVNV;
1286 assert(new_type > SVt_PV);
1287 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1288 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1295 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1296 there's no way that it can be safely upgraded, because perl.c
1297 expects to Safefree(SvANY(PL_mess_sv)) */
1298 assert(sv != PL_mess_sv);
1301 if (UNLIKELY(old_type_details->cant_upgrade))
1302 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1303 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1306 if (UNLIKELY(old_type > new_type))
1307 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1308 (int)old_type, (int)new_type);
1310 new_type_details = bodies_by_type + new_type;
1312 SvFLAGS(sv) &= ~SVTYPEMASK;
1313 SvFLAGS(sv) |= new_type;
1315 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1316 the return statements above will have triggered. */
1317 assert (new_type != SVt_NULL);
1320 assert(old_type == SVt_NULL);
1321 SET_SVANY_FOR_BODYLESS_IV(sv);
1325 assert(old_type == SVt_NULL);
1326 #if NVSIZE <= IVSIZE
1327 SET_SVANY_FOR_BODYLESS_NV(sv);
1329 SvANY(sv) = new_XNV();
1335 assert(new_type_details->body_size);
1338 assert(new_type_details->arena);
1339 assert(new_type_details->arena_size);
1340 /* This points to the start of the allocated area. */
1341 new_body_inline(new_body, new_type);
1342 Zero(new_body, new_type_details->body_size, char);
1343 new_body = ((char *)new_body) - new_type_details->offset;
1345 /* We always allocated the full length item with PURIFY. To do this
1346 we fake things so that arena is false for all 16 types.. */
1347 new_body = new_NOARENAZ(new_type_details);
1349 SvANY(sv) = new_body;
1350 if (new_type == SVt_PVAV) {
1354 if (old_type_details->body_size) {
1357 /* It will have been zeroed when the new body was allocated.
1358 Lets not write to it, in case it confuses a write-back
1364 #ifndef NODEFAULT_SHAREKEYS
1365 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1367 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1368 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1371 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1372 The target created by newSVrv also is, and it can have magic.
1373 However, it never has SvPVX set.
1375 if (old_type == SVt_IV) {
1377 } else if (old_type >= SVt_PV) {
1378 assert(SvPVX_const(sv) == 0);
1381 if (old_type >= SVt_PVMG) {
1382 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1383 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1385 sv->sv_u.svu_array = NULL; /* or svu_hash */
1390 /* XXX Is this still needed? Was it ever needed? Surely as there is
1391 no route from NV to PVIV, NOK can never be true */
1392 assert(!SvNOKp(sv));
1406 assert(new_type_details->body_size);
1407 /* We always allocated the full length item with PURIFY. To do this
1408 we fake things so that arena is false for all 16 types.. */
1409 if(new_type_details->arena) {
1410 /* This points to the start of the allocated area. */
1411 new_body_inline(new_body, new_type);
1412 Zero(new_body, new_type_details->body_size, char);
1413 new_body = ((char *)new_body) - new_type_details->offset;
1415 new_body = new_NOARENAZ(new_type_details);
1417 SvANY(sv) = new_body;
1419 if (old_type_details->copy) {
1420 /* There is now the potential for an upgrade from something without
1421 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1422 int offset = old_type_details->offset;
1423 int length = old_type_details->copy;
1425 if (new_type_details->offset > old_type_details->offset) {
1426 const int difference
1427 = new_type_details->offset - old_type_details->offset;
1428 offset += difference;
1429 length -= difference;
1431 assert (length >= 0);
1433 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1437 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1438 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1439 * correct 0.0 for us. Otherwise, if the old body didn't have an
1440 * NV slot, but the new one does, then we need to initialise the
1441 * freshly created NV slot with whatever the correct bit pattern is
1443 if (old_type_details->zero_nv && !new_type_details->zero_nv
1444 && !isGV_with_GP(sv))
1448 if (UNLIKELY(new_type == SVt_PVIO)) {
1449 IO * const io = MUTABLE_IO(sv);
1450 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1453 /* Clear the stashcache because a new IO could overrule a package
1455 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1456 hv_clear(PL_stashcache);
1458 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1459 IoPAGE_LEN(sv) = 60;
1461 if (old_type < SVt_PV) {
1462 /* referent will be NULL unless the old type was SVt_IV emulating
1464 sv->sv_u.svu_rv = referent;
1468 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1469 (unsigned long)new_type);
1472 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1473 and sometimes SVt_NV */
1474 if (old_type_details->body_size) {
1478 /* Note that there is an assumption that all bodies of types that
1479 can be upgraded came from arenas. Only the more complex non-
1480 upgradable types are allowed to be directly malloc()ed. */
1481 assert(old_type_details->arena);
1482 del_body((void*)((char*)old_body + old_type_details->offset),
1483 &PL_body_roots[old_type]);
1489 =for apidoc sv_backoff
1491 Remove any string offset. You should normally use the C<SvOOK_off> macro
1497 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1498 prior to 5.23.4 this function always returned 0
1502 Perl_sv_backoff(SV *const sv)
1505 const char * const s = SvPVX_const(sv);
1507 PERL_ARGS_ASSERT_SV_BACKOFF;
1510 assert(SvTYPE(sv) != SVt_PVHV);
1511 assert(SvTYPE(sv) != SVt_PVAV);
1513 SvOOK_offset(sv, delta);
1515 SvLEN_set(sv, SvLEN(sv) + delta);
1516 SvPV_set(sv, SvPVX(sv) - delta);
1517 SvFLAGS(sv) &= ~SVf_OOK;
1518 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1523 /* forward declaration */
1524 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1530 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1531 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1532 Use the C<SvGROW> wrapper instead.
1539 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1543 PERL_ARGS_ASSERT_SV_GROW;
1547 if (SvTYPE(sv) < SVt_PV) {
1548 sv_upgrade(sv, SVt_PV);
1549 s = SvPVX_mutable(sv);
1551 else if (SvOOK(sv)) { /* pv is offset? */
1553 s = SvPVX_mutable(sv);
1554 if (newlen > SvLEN(sv))
1555 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1559 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1560 s = SvPVX_mutable(sv);
1563 #ifdef PERL_COPY_ON_WRITE
1564 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1565 * to store the COW count. So in general, allocate one more byte than
1566 * asked for, to make it likely this byte is always spare: and thus
1567 * make more strings COW-able.
1569 * Only increment if the allocation isn't MEM_SIZE_MAX,
1570 * otherwise it will wrap to 0.
1572 if ( newlen != MEM_SIZE_MAX )
1576 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1577 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1580 if (newlen > SvLEN(sv)) { /* need more room? */
1581 STRLEN minlen = SvCUR(sv);
1582 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1583 if (newlen < minlen)
1585 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1587 /* Don't round up on the first allocation, as odds are pretty good that
1588 * the initial request is accurate as to what is really needed */
1590 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1591 if (rounded > newlen)
1595 if (SvLEN(sv) && s) {
1596 s = (char*)saferealloc(s, newlen);
1599 s = (char*)safemalloc(newlen);
1600 if (SvPVX_const(sv) && SvCUR(sv)) {
1601 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1605 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1606 /* Do this here, do it once, do it right, and then we will never get
1607 called back into sv_grow() unless there really is some growing
1609 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1611 SvLEN_set(sv, newlen);
1618 =for apidoc sv_setiv
1620 Copies an integer into the given SV, upgrading first if necessary.
1621 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1627 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1629 PERL_ARGS_ASSERT_SV_SETIV;
1631 SV_CHECK_THINKFIRST_COW_DROP(sv);
1632 switch (SvTYPE(sv)) {
1635 sv_upgrade(sv, SVt_IV);
1638 sv_upgrade(sv, SVt_PVIV);
1642 if (!isGV_with_GP(sv))
1650 /* diag_listed_as: Can't coerce %s to %s in %s */
1651 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1653 NOT_REACHED; /* NOTREACHED */
1657 (void)SvIOK_only(sv); /* validate number */
1663 =for apidoc sv_setiv_mg
1665 Like C<sv_setiv>, but also handles 'set' magic.
1671 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1680 =for apidoc sv_setuv
1682 Copies an unsigned integer into the given SV, upgrading first if necessary.
1683 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1689 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1691 PERL_ARGS_ASSERT_SV_SETUV;
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1714 =for apidoc sv_setuv_mg
1716 Like C<sv_setuv>, but also handles 'set' magic.
1722 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1731 =for apidoc sv_setnv
1733 Copies a double into the given SV, upgrading first if necessary.
1734 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1740 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1742 PERL_ARGS_ASSERT_SV_SETNV;
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
1748 sv_upgrade(sv, SVt_NV);
1752 sv_upgrade(sv, SVt_PVNV);
1756 if (!isGV_with_GP(sv))
1764 /* diag_listed_as: Can't coerce %s to %s in %s */
1765 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1767 NOT_REACHED; /* NOTREACHED */
1772 (void)SvNOK_only(sv); /* validate number */
1777 =for apidoc sv_setnv_mg
1779 Like C<sv_setnv>, but also handles 'set' magic.
1785 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1787 PERL_ARGS_ASSERT_SV_SETNV_MG;
1793 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1794 * not incrementable warning display.
1795 * Originally part of S_not_a_number().
1796 * The return value may be != tmpbuf.
1800 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1803 PERL_ARGS_ASSERT_SV_DISPLAY;
1806 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1807 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1810 const char * const limit = tmpbuf + tmpbuf_size - 8;
1811 /* each *s can expand to 4 chars + "...\0",
1812 i.e. need room for 8 chars */
1814 const char *s = SvPVX_const(sv);
1815 const char * const end = s + SvCUR(sv);
1816 for ( ; s < end && d < limit; s++ ) {
1818 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1822 /* Map to ASCII "equivalent" of Latin1 */
1823 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1829 else if (ch == '\r') {
1833 else if (ch == '\f') {
1837 else if (ch == '\\') {
1841 else if (ch == '\0') {
1845 else if (isPRINT_LC(ch))
1864 /* Print an "isn't numeric" warning, using a cleaned-up,
1865 * printable version of the offending string
1869 S_not_a_number(pTHX_ SV *const sv)
1874 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1876 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1879 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1880 /* diag_listed_as: Argument "%s" isn't numeric%s */
1881 "Argument \"%s\" isn't numeric in %s", pv,
1884 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1885 /* diag_listed_as: Argument "%s" isn't numeric%s */
1886 "Argument \"%s\" isn't numeric", pv);
1890 S_not_incrementable(pTHX_ SV *const sv) {
1894 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1896 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1898 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1899 "Argument \"%s\" treated as 0 in increment (++)", pv);
1903 =for apidoc looks_like_number
1905 Test if the content of an SV looks like a number (or is a number).
1906 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1907 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1914 Perl_looks_like_number(pTHX_ SV *const sv)
1920 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1922 if (SvPOK(sv) || SvPOKp(sv)) {
1923 sbegin = SvPV_nomg_const(sv, len);
1926 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1927 numtype = grok_number(sbegin, len, NULL);
1928 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1932 S_glob_2number(pTHX_ GV * const gv)
1934 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1936 /* We know that all GVs stringify to something that is not-a-number,
1937 so no need to test that. */
1938 if (ckWARN(WARN_NUMERIC))
1940 SV *const buffer = sv_newmortal();
1941 gv_efullname3(buffer, gv, "*");
1942 not_a_number(buffer);
1944 /* We just want something true to return, so that S_sv_2iuv_common
1945 can tail call us and return true. */
1949 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1950 until proven guilty, assume that things are not that bad... */
1955 As 64 bit platforms often have an NV that doesn't preserve all bits of
1956 an IV (an assumption perl has been based on to date) it becomes necessary
1957 to remove the assumption that the NV always carries enough precision to
1958 recreate the IV whenever needed, and that the NV is the canonical form.
1959 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1960 precision as a side effect of conversion (which would lead to insanity
1961 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1962 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1963 where precision was lost, and IV/UV/NV slots that have a valid conversion
1964 which has lost no precision
1965 2) to ensure that if a numeric conversion to one form is requested that
1966 would lose precision, the precise conversion (or differently
1967 imprecise conversion) is also performed and cached, to prevent
1968 requests for different numeric formats on the same SV causing
1969 lossy conversion chains. (lossless conversion chains are perfectly
1974 SvIOKp is true if the IV slot contains a valid value
1975 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1976 SvNOKp is true if the NV slot contains a valid value
1977 SvNOK is true only if the NV value is accurate
1980 while converting from PV to NV, check to see if converting that NV to an
1981 IV(or UV) would lose accuracy over a direct conversion from PV to
1982 IV(or UV). If it would, cache both conversions, return NV, but mark
1983 SV as IOK NOKp (ie not NOK).
1985 While converting from PV to IV, check to see if converting that IV to an
1986 NV would lose accuracy over a direct conversion from PV to NV. If it
1987 would, cache both conversions, flag similarly.
1989 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1990 correctly because if IV & NV were set NV *always* overruled.
1991 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1992 changes - now IV and NV together means that the two are interchangeable:
1993 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1995 The benefit of this is that operations such as pp_add know that if
1996 SvIOK is true for both left and right operands, then integer addition
1997 can be used instead of floating point (for cases where the result won't
1998 overflow). Before, floating point was always used, which could lead to
1999 loss of precision compared with integer addition.
2001 * making IV and NV equal status should make maths accurate on 64 bit
2003 * may speed up maths somewhat if pp_add and friends start to use
2004 integers when possible instead of fp. (Hopefully the overhead in
2005 looking for SvIOK and checking for overflow will not outweigh the
2006 fp to integer speedup)
2007 * will slow down integer operations (callers of SvIV) on "inaccurate"
2008 values, as the change from SvIOK to SvIOKp will cause a call into
2009 sv_2iv each time rather than a macro access direct to the IV slot
2010 * should speed up number->string conversion on integers as IV is
2011 favoured when IV and NV are equally accurate
2013 ####################################################################
2014 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2015 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2016 On the other hand, SvUOK is true iff UV.
2017 ####################################################################
2019 Your mileage will vary depending your CPU's relative fp to integer
2023 #ifndef NV_PRESERVES_UV
2024 # define IS_NUMBER_UNDERFLOW_IV 1
2025 # define IS_NUMBER_UNDERFLOW_UV 2
2026 # define IS_NUMBER_IV_AND_UV 2
2027 # define IS_NUMBER_OVERFLOW_IV 4
2028 # define IS_NUMBER_OVERFLOW_UV 5
2030 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2032 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2034 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2040 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2041 PERL_UNUSED_CONTEXT;
2043 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));
2044 if (SvNVX(sv) < (NV)IV_MIN) {
2045 (void)SvIOKp_on(sv);
2047 SvIV_set(sv, IV_MIN);
2048 return IS_NUMBER_UNDERFLOW_IV;
2050 if (SvNVX(sv) > (NV)UV_MAX) {
2051 (void)SvIOKp_on(sv);
2054 SvUV_set(sv, UV_MAX);
2055 return IS_NUMBER_OVERFLOW_UV;
2057 (void)SvIOKp_on(sv);
2059 /* Can't use strtol etc to convert this string. (See truth table in
2061 if (SvNVX(sv) <= (UV)IV_MAX) {
2062 SvIV_set(sv, I_V(SvNVX(sv)));
2063 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2064 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2066 /* Integer is imprecise. NOK, IOKp */
2068 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2071 SvUV_set(sv, U_V(SvNVX(sv)));
2072 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2073 if (SvUVX(sv) == UV_MAX) {
2074 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2075 possibly be preserved by NV. Hence, it must be overflow.
2077 return IS_NUMBER_OVERFLOW_UV;
2079 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2081 /* Integer is imprecise. NOK, IOKp */
2083 return IS_NUMBER_OVERFLOW_IV;
2085 #endif /* !NV_PRESERVES_UV*/
2087 /* If numtype is infnan, set the NV of the sv accordingly.
2088 * If numtype is anything else, try setting the NV using Atof(PV). */
2090 # pragma warning(push)
2091 # pragma warning(disable:4756;disable:4056)
2094 S_sv_setnv(pTHX_ SV* sv, int numtype)
2096 bool pok = cBOOL(SvPOK(sv));
2099 if ((numtype & IS_NUMBER_INFINITY)) {
2100 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2105 if ((numtype & IS_NUMBER_NAN)) {
2106 SvNV_set(sv, NV_NAN);
2111 SvNV_set(sv, Atof(SvPVX_const(sv)));
2112 /* Purposefully no true nok here, since we don't want to blow
2113 * away the possible IOK/UV of an existing sv. */
2116 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2118 SvPOK_on(sv); /* PV is okay, though. */
2122 # pragma warning(pop)
2126 S_sv_2iuv_common(pTHX_ SV *const sv)
2128 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2131 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2132 * without also getting a cached IV/UV from it at the same time
2133 * (ie PV->NV conversion should detect loss of accuracy and cache
2134 * IV or UV at same time to avoid this. */
2135 /* IV-over-UV optimisation - choose to cache IV if possible */
2137 if (SvTYPE(sv) == SVt_NV)
2138 sv_upgrade(sv, SVt_PVNV);
2140 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2141 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2142 certainly cast into the IV range at IV_MAX, whereas the correct
2143 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2145 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2146 if (Perl_isnan(SvNVX(sv))) {
2152 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2153 SvIV_set(sv, I_V(SvNVX(sv)));
2154 if (SvNVX(sv) == (NV) SvIVX(sv)
2155 #ifndef NV_PRESERVES_UV
2156 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2157 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2158 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2159 /* Don't flag it as "accurately an integer" if the number
2160 came from a (by definition imprecise) NV operation, and
2161 we're outside the range of NV integer precision */
2165 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2167 /* scalar has trailing garbage, eg "42a" */
2169 DEBUG_c(PerlIO_printf(Perl_debug_log,
2170 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2176 /* IV not precise. No need to convert from PV, as NV
2177 conversion would already have cached IV if it detected
2178 that PV->IV would be better than PV->NV->IV
2179 flags already correct - don't set public IOK. */
2180 DEBUG_c(PerlIO_printf(Perl_debug_log,
2181 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2186 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2187 but the cast (NV)IV_MIN rounds to a the value less (more
2188 negative) than IV_MIN which happens to be equal to SvNVX ??
2189 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2190 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2191 (NV)UVX == NVX are both true, but the values differ. :-(
2192 Hopefully for 2s complement IV_MIN is something like
2193 0x8000000000000000 which will be exact. NWC */
2196 SvUV_set(sv, U_V(SvNVX(sv)));
2198 (SvNVX(sv) == (NV) SvUVX(sv))
2199 #ifndef NV_PRESERVES_UV
2200 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2201 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2202 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2203 /* Don't flag it as "accurately an integer" if the number
2204 came from a (by definition imprecise) NV operation, and
2205 we're outside the range of NV integer precision */
2211 DEBUG_c(PerlIO_printf(Perl_debug_log,
2212 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2218 else if (SvPOKp(sv)) {
2221 const char *s = SvPVX_const(sv);
2222 const STRLEN cur = SvCUR(sv);
2224 /* short-cut for a single digit string like "1" */
2229 if (SvTYPE(sv) < SVt_PVIV)
2230 sv_upgrade(sv, SVt_PVIV);
2232 SvIV_set(sv, (IV)(c - '0'));
2237 numtype = grok_number(s, cur, &value);
2238 /* We want to avoid a possible problem when we cache an IV/ a UV which
2239 may be later translated to an NV, and the resulting NV is not
2240 the same as the direct translation of the initial string
2241 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2242 be careful to ensure that the value with the .456 is around if the
2243 NV value is requested in the future).
2245 This means that if we cache such an IV/a UV, we need to cache the
2246 NV as well. Moreover, we trade speed for space, and do not
2247 cache the NV if we are sure it's not needed.
2250 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2251 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2252 == IS_NUMBER_IN_UV) {
2253 /* It's definitely an integer, only upgrade to PVIV */
2254 if (SvTYPE(sv) < SVt_PVIV)
2255 sv_upgrade(sv, SVt_PVIV);
2257 } else if (SvTYPE(sv) < SVt_PVNV)
2258 sv_upgrade(sv, SVt_PVNV);
2260 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2261 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2263 S_sv_setnv(aTHX_ sv, numtype);
2267 /* If NVs preserve UVs then we only use the UV value if we know that
2268 we aren't going to call atof() below. If NVs don't preserve UVs
2269 then the value returned may have more precision than atof() will
2270 return, even though value isn't perfectly accurate. */
2271 if ((numtype & (IS_NUMBER_IN_UV
2272 #ifdef NV_PRESERVES_UV
2275 )) == IS_NUMBER_IN_UV) {
2276 /* This won't turn off the public IOK flag if it was set above */
2277 (void)SvIOKp_on(sv);
2279 if (!(numtype & IS_NUMBER_NEG)) {
2281 if (value <= (UV)IV_MAX) {
2282 SvIV_set(sv, (IV)value);
2284 /* it didn't overflow, and it was positive. */
2285 SvUV_set(sv, value);
2289 /* 2s complement assumption */
2290 if (value <= (UV)IV_MIN) {
2291 SvIV_set(sv, value == (UV)IV_MIN
2292 ? IV_MIN : -(IV)value);
2294 /* Too negative for an IV. This is a double upgrade, but
2295 I'm assuming it will be rare. */
2296 if (SvTYPE(sv) < SVt_PVNV)
2297 sv_upgrade(sv, SVt_PVNV);
2301 SvNV_set(sv, -(NV)value);
2302 SvIV_set(sv, IV_MIN);
2306 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2307 will be in the previous block to set the IV slot, and the next
2308 block to set the NV slot. So no else here. */
2310 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2311 != IS_NUMBER_IN_UV) {
2312 /* It wasn't an (integer that doesn't overflow the UV). */
2313 S_sv_setnv(aTHX_ sv, numtype);
2315 if (! numtype && ckWARN(WARN_NUMERIC))
2318 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2319 PTR2UV(sv), SvNVX(sv)));
2321 #ifdef NV_PRESERVES_UV
2322 (void)SvIOKp_on(sv);
2324 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2325 if (Perl_isnan(SvNVX(sv))) {
2331 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2332 SvIV_set(sv, I_V(SvNVX(sv)));
2333 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2336 NOOP; /* Integer is imprecise. NOK, IOKp */
2338 /* UV will not work better than IV */
2340 if (SvNVX(sv) > (NV)UV_MAX) {
2342 /* Integer is inaccurate. NOK, IOKp, is UV */
2343 SvUV_set(sv, UV_MAX);
2345 SvUV_set(sv, U_V(SvNVX(sv)));
2346 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2347 NV preservse UV so can do correct comparison. */
2348 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2351 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2356 #else /* NV_PRESERVES_UV */
2357 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2358 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2359 /* The IV/UV slot will have been set from value returned by
2360 grok_number above. The NV slot has just been set using
2363 assert (SvIOKp(sv));
2365 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2366 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2367 /* Small enough to preserve all bits. */
2368 (void)SvIOKp_on(sv);
2370 SvIV_set(sv, I_V(SvNVX(sv)));
2371 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2373 /* Assumption: first non-preserved integer is < IV_MAX,
2374 this NV is in the preserved range, therefore: */
2375 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2377 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);
2381 0 0 already failed to read UV.
2382 0 1 already failed to read UV.
2383 1 0 you won't get here in this case. IV/UV
2384 slot set, public IOK, Atof() unneeded.
2385 1 1 already read UV.
2386 so there's no point in sv_2iuv_non_preserve() attempting
2387 to use atol, strtol, strtoul etc. */
2389 sv_2iuv_non_preserve (sv, numtype);
2391 sv_2iuv_non_preserve (sv);
2395 #endif /* NV_PRESERVES_UV */
2396 /* It might be more code efficient to go through the entire logic above
2397 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2398 gets complex and potentially buggy, so more programmer efficient
2399 to do it this way, by turning off the public flags: */
2401 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2405 if (isGV_with_GP(sv))
2406 return glob_2number(MUTABLE_GV(sv));
2408 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2410 if (SvTYPE(sv) < SVt_IV)
2411 /* Typically the caller expects that sv_any is not NULL now. */
2412 sv_upgrade(sv, SVt_IV);
2413 /* Return 0 from the caller. */
2420 =for apidoc sv_2iv_flags
2422 Return the integer value of an SV, doing any necessary string
2423 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2424 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2430 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2432 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2434 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2435 && SvTYPE(sv) != SVt_PVFM);
2437 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2443 if (flags & SV_SKIP_OVERLOAD)
2445 tmpstr = AMG_CALLunary(sv, numer_amg);
2446 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2447 return SvIV(tmpstr);
2450 return PTR2IV(SvRV(sv));
2453 if (SvVALID(sv) || isREGEXP(sv)) {
2454 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2455 must not let them cache IVs.
2456 In practice they are extremely unlikely to actually get anywhere
2457 accessible by user Perl code - the only way that I'm aware of is when
2458 a constant subroutine which is used as the second argument to index.
2460 Regexps have no SvIVX and SvNVX fields.
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2468 = grok_number(ptr, SvCUR(sv), &value);
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (numtype & IS_NUMBER_NEG) {
2474 if (value < (UV)IV_MIN)
2477 if (value < (UV)IV_MAX)
2482 /* Quite wrong but no good choices. */
2483 if ((numtype & IS_NUMBER_INFINITY)) {
2484 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2485 } else if ((numtype & IS_NUMBER_NAN)) {
2486 return 0; /* So wrong. */
2490 if (ckWARN(WARN_NUMERIC))
2493 return I_V(Atof(ptr));
2497 if (SvTHINKFIRST(sv)) {
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2506 if (S_sv_2iuv_common(aTHX_ sv))
2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2511 PTR2UV(sv),SvIVX(sv)));
2512 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2516 =for apidoc sv_2uv_flags
2518 Return the unsigned integer value of an SV, doing any necessary string
2519 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2520 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2526 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2528 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2530 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2536 if (flags & SV_SKIP_OVERLOAD)
2538 tmpstr = AMG_CALLunary(sv, numer_amg);
2539 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2540 return SvUV(tmpstr);
2543 return PTR2UV(SvRV(sv));
2546 if (SvVALID(sv) || isREGEXP(sv)) {
2547 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2548 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2549 Regexps have no SvIVX and SvNVX fields. */
2553 const char * const ptr =
2554 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2556 = grok_number(ptr, SvCUR(sv), &value);
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
2560 /* It's definitely an integer */
2561 if (!(numtype & IS_NUMBER_NEG))
2565 /* Quite wrong but no good choices. */
2566 if ((numtype & IS_NUMBER_INFINITY)) {
2567 return UV_MAX; /* So wrong. */
2568 } else if ((numtype & IS_NUMBER_NAN)) {
2569 return 0; /* So wrong. */
2573 if (ckWARN(WARN_NUMERIC))
2576 return U_V(Atof(ptr));
2580 if (SvTHINKFIRST(sv)) {
2581 if (SvREADONLY(sv) && !SvOK(sv)) {
2582 if (ckWARN(WARN_UNINITIALIZED))
2589 if (S_sv_2iuv_common(aTHX_ sv))
2593 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2594 PTR2UV(sv),SvUVX(sv)));
2595 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2599 =for apidoc sv_2nv_flags
2601 Return the num value of an SV, doing any necessary string or integer
2602 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2603 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2609 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2611 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2613 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2614 && SvTYPE(sv) != SVt_PVFM);
2615 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2616 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2617 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2618 Regexps have no SvIVX and SvNVX fields. */
2620 if (flags & SV_GMAGIC)
2624 if (SvPOKp(sv) && !SvIOKp(sv)) {
2625 ptr = SvPVX_const(sv);
2626 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2627 !grok_number(ptr, SvCUR(sv), NULL))
2633 return (NV)SvUVX(sv);
2635 return (NV)SvIVX(sv);
2640 assert(SvTYPE(sv) >= SVt_PVMG);
2641 /* This falls through to the report_uninit near the end of the
2643 } else if (SvTHINKFIRST(sv)) {
2648 if (flags & SV_SKIP_OVERLOAD)
2650 tmpstr = AMG_CALLunary(sv, numer_amg);
2651 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2652 return SvNV(tmpstr);
2655 return PTR2NV(SvRV(sv));
2657 if (SvREADONLY(sv) && !SvOK(sv)) {
2658 if (ckWARN(WARN_UNINITIALIZED))
2663 if (SvTYPE(sv) < SVt_NV) {
2664 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2665 sv_upgrade(sv, SVt_NV);
2666 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2668 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2669 STORE_LC_NUMERIC_SET_STANDARD();
2670 PerlIO_printf(Perl_debug_log,
2671 "0x%" UVxf " num(%" NVgf ")\n",
2672 PTR2UV(sv), SvNVX(sv));
2673 RESTORE_LC_NUMERIC();
2675 CLANG_DIAG_RESTORE_STMT;
2678 else if (SvTYPE(sv) < SVt_PVNV)
2679 sv_upgrade(sv, SVt_PVNV);
2684 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2685 #ifdef NV_PRESERVES_UV
2691 /* Only set the public NV OK flag if this NV preserves the IV */
2692 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2694 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2695 : (SvIVX(sv) == I_V(SvNVX(sv))))
2701 else if (SvPOKp(sv)) {
2703 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2704 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2706 #ifdef NV_PRESERVES_UV
2707 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2708 == IS_NUMBER_IN_UV) {
2709 /* It's definitely an integer */
2710 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2712 S_sv_setnv(aTHX_ sv, numtype);
2719 SvNV_set(sv, Atof(SvPVX_const(sv)));
2720 /* Only set the public NV OK flag if this NV preserves the value in
2721 the PV at least as well as an IV/UV would.
2722 Not sure how to do this 100% reliably. */
2723 /* if that shift count is out of range then Configure's test is
2724 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2726 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2727 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2728 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2729 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2730 /* Can't use strtol etc to convert this string, so don't try.
2731 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2734 /* value has been set. It may not be precise. */
2735 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2736 /* 2s complement assumption for (UV)IV_MIN */
2737 SvNOK_on(sv); /* Integer is too negative. */
2742 if (numtype & IS_NUMBER_NEG) {
2743 /* -IV_MIN is undefined, but we should never reach
2744 * this point with both IS_NUMBER_NEG and value ==
2746 assert(value != (UV)IV_MIN);
2747 SvIV_set(sv, -(IV)value);
2748 } else if (value <= (UV)IV_MAX) {
2749 SvIV_set(sv, (IV)value);
2751 SvUV_set(sv, value);
2755 if (numtype & IS_NUMBER_NOT_INT) {
2756 /* I believe that even if the original PV had decimals,
2757 they are lost beyond the limit of the FP precision.
2758 However, neither is canonical, so both only get p
2759 flags. NWC, 2000/11/25 */
2760 /* Both already have p flags, so do nothing */
2762 const NV nv = SvNVX(sv);
2763 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2764 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2765 if (SvIVX(sv) == I_V(nv)) {
2768 /* It had no "." so it must be integer. */
2772 /* between IV_MAX and NV(UV_MAX).
2773 Could be slightly > UV_MAX */
2775 if (numtype & IS_NUMBER_NOT_INT) {
2776 /* UV and NV both imprecise. */
2778 const UV nv_as_uv = U_V(nv);
2780 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2789 /* It might be more code efficient to go through the entire logic above
2790 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2791 gets complex and potentially buggy, so more programmer efficient
2792 to do it this way, by turning off the public flags: */
2794 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2795 #endif /* NV_PRESERVES_UV */
2798 if (isGV_with_GP(sv)) {
2799 glob_2number(MUTABLE_GV(sv));
2803 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2805 assert (SvTYPE(sv) >= SVt_NV);
2806 /* Typically the caller expects that sv_any is not NULL now. */
2807 /* XXX Ilya implies that this is a bug in callers that assume this
2808 and ideally should be fixed. */
2811 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2813 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2814 STORE_LC_NUMERIC_SET_STANDARD();
2815 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2816 PTR2UV(sv), SvNVX(sv));
2817 RESTORE_LC_NUMERIC();
2819 CLANG_DIAG_RESTORE_STMT;
2826 Return an SV with the numeric value of the source SV, doing any necessary
2827 reference or overload conversion. The caller is expected to have handled
2834 Perl_sv_2num(pTHX_ SV *const sv)
2836 PERL_ARGS_ASSERT_SV_2NUM;
2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2849 /* int2str_table: lookup table containing string representations of all
2850 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2851 * int2str_table.arr[12*2] is "12".
2853 * We are going to read two bytes at a time, so we have to ensure that
2854 * the array is aligned to a 2 byte boundary. That's why it was made a
2855 * union with a dummy U16 member. */
2856 static const union {
2859 } int2str_table = {{
2860 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2861 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2862 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2863 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2864 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2865 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2866 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2867 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2868 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2869 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2870 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2871 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2872 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2873 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2877 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2878 * UV as a string towards the end of buf, and return pointers to start and
2881 * We assume that buf is at least TYPE_CHARS(UV) long.
2884 PERL_STATIC_INLINE char *
2885 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2887 char *ptr = buf + TYPE_CHARS(UV);
2888 char * const ebuf = ptr;
2890 U16 *word_ptr, *word_table;
2892 PERL_ARGS_ASSERT_UIV_2BUF;
2894 /* ptr has to be properly aligned, because we will cast it to U16* */
2895 assert(PTR2nat(ptr) % 2 == 0);
2896 /* we are going to read/write two bytes at a time */
2897 word_ptr = (U16*)ptr;
2898 word_table = (U16*)int2str_table.arr;
2900 if (UNLIKELY(is_uv))
2906 /* Using 0- here to silence bogus warning from MS VC */
2907 uv = (UV) (0 - (UV) iv);
2912 *--word_ptr = word_table[uv % 100];
2915 ptr = (char*)word_ptr;
2918 *--ptr = (char)uv + '0';
2920 *--word_ptr = word_table[uv];
2921 ptr = (char*)word_ptr;
2931 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2932 * infinity or a not-a-number, writes the appropriate strings to the
2933 * buffer, including a zero byte. On success returns the written length,
2934 * excluding the zero byte, on failure (not an infinity, not a nan)
2935 * returns zero, assert-fails on maxlen being too short.
2937 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2938 * shared string constants we point to, instead of generating a new
2939 * string for each instance. */
2941 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2943 assert(maxlen >= 4);
2944 if (Perl_isinf(nv)) {
2946 if (maxlen < 5) /* "-Inf\0" */
2956 else if (Perl_isnan(nv)) {
2960 /* XXX optionally output the payload mantissa bits as
2961 * "(unsigned)" (to match the nan("...") C99 function,
2962 * or maybe as "(0xhhh...)" would make more sense...
2963 * provide a format string so that the user can decide?
2964 * NOTE: would affect the maxlen and assert() logic.*/
2969 assert((s == buffer + 3) || (s == buffer + 4));
2975 =for apidoc sv_2pv_flags
2977 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2978 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2979 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2980 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2986 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2990 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2992 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2993 && SvTYPE(sv) != SVt_PVFM);
2994 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2999 if (flags & SV_SKIP_OVERLOAD)
3001 tmpstr = AMG_CALLunary(sv, string_amg);
3002 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3003 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3005 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3009 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3010 if (flags & SV_CONST_RETURN) {
3011 pv = (char *) SvPVX_const(tmpstr);
3013 pv = (flags & SV_MUTABLE_RETURN)
3014 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3017 *lp = SvCUR(tmpstr);
3019 pv = sv_2pv_flags(tmpstr, lp, flags);
3032 SV *const referent = SvRV(sv);
3036 retval = buffer = savepvn("NULLREF", len);
3037 } else if (SvTYPE(referent) == SVt_REGEXP &&
3038 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3039 amagic_is_enabled(string_amg))) {
3040 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3044 /* If the regex is UTF-8 we want the containing scalar to
3045 have an UTF-8 flag too */
3052 *lp = RX_WRAPLEN(re);
3054 return RX_WRAPPED(re);
3056 const char *const typestr = sv_reftype(referent, 0);
3057 const STRLEN typelen = strlen(typestr);
3058 UV addr = PTR2UV(referent);
3059 const char *stashname = NULL;
3060 STRLEN stashnamelen = 0; /* hush, gcc */
3061 const char *buffer_end;
3063 if (SvOBJECT(referent)) {
3064 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3067 stashname = HEK_KEY(name);
3068 stashnamelen = HEK_LEN(name);
3070 if (HEK_UTF8(name)) {
3076 stashname = "__ANON__";
3079 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3080 + 2 * sizeof(UV) + 2 /* )\0 */;
3082 len = typelen + 3 /* (0x */
3083 + 2 * sizeof(UV) + 2 /* )\0 */;
3086 Newx(buffer, len, char);
3087 buffer_end = retval = buffer + len;
3089 /* Working backwards */
3093 *--retval = PL_hexdigit[addr & 15];
3094 } while (addr >>= 4);
3100 memcpy(retval, typestr, typelen);
3104 retval -= stashnamelen;
3105 memcpy(retval, stashname, stashnamelen);
3107 /* retval may not necessarily have reached the start of the
3109 assert (retval >= buffer);
3111 len = buffer_end - retval - 1; /* -1 for that \0 */
3123 if (flags & SV_MUTABLE_RETURN)
3124 return SvPVX_mutable(sv);
3125 if (flags & SV_CONST_RETURN)
3126 return (char *)SvPVX_const(sv);
3131 /* I'm assuming that if both IV and NV are equally valid then
3132 converting the IV is going to be more efficient */
3133 const U32 isUIOK = SvIsUV(sv);
3134 /* The purpose of this union is to ensure that arr is aligned on
3135 a 2 byte boundary, because that is what uiv_2buf() requires */
3137 char arr[TYPE_CHARS(UV)];
3143 if (SvTYPE(sv) < SVt_PVIV)
3144 sv_upgrade(sv, SVt_PVIV);
3145 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3147 /* inlined from sv_setpvn */
3148 s = SvGROW_mutable(sv, len + 1);
3149 Move(ptr, s, len, char);
3154 else if (SvNOK(sv)) {
3155 if (SvTYPE(sv) < SVt_PVNV)
3156 sv_upgrade(sv, SVt_PVNV);
3157 if (SvNVX(sv) == 0.0
3158 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3159 && !Perl_isnan(SvNVX(sv))
3162 s = SvGROW_mutable(sv, 2);
3167 STRLEN size = 5; /* "-Inf\0" */
3169 s = SvGROW_mutable(sv, size);
3170 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3176 /* some Xenix systems wipe out errno here */
3185 5 + /* exponent digits */
3189 s = SvGROW_mutable(sv, size);
3190 #ifndef USE_LOCALE_NUMERIC
3191 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3197 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3198 STORE_LC_NUMERIC_SET_TO_NEEDED();
3200 local_radix = _NOT_IN_NUMERIC_STANDARD;
3201 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3202 size += SvCUR(PL_numeric_radix_sv) - 1;
3203 s = SvGROW_mutable(sv, size);
3206 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3208 /* If the radix character is UTF-8, and actually is in the
3209 * output, turn on the UTF-8 flag for the scalar */
3211 && SvUTF8(PL_numeric_radix_sv)
3212 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3217 RESTORE_LC_NUMERIC();
3220 /* We don't call SvPOK_on(), because it may come to
3221 * pass that the locale changes so that the
3222 * stringification we just did is no longer correct. We
3223 * will have to re-stringify every time it is needed */
3230 else if (isGV_with_GP(sv)) {
3231 GV *const gv = MUTABLE_GV(sv);
3232 SV *const buffer = sv_newmortal();
3234 gv_efullname3(buffer, gv, "*");
3236 assert(SvPOK(buffer));
3242 *lp = SvCUR(buffer);
3243 return SvPVX(buffer);
3248 if (flags & SV_UNDEF_RETURNS_NULL)
3250 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3252 /* Typically the caller expects that sv_any is not NULL now. */
3253 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3254 sv_upgrade(sv, SVt_PV);
3259 const STRLEN len = s - SvPVX_const(sv);
3264 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3265 PTR2UV(sv),SvPVX_const(sv)));
3266 if (flags & SV_CONST_RETURN)
3267 return (char *)SvPVX_const(sv);
3268 if (flags & SV_MUTABLE_RETURN)
3269 return SvPVX_mutable(sv);
3274 =for apidoc sv_copypv
3276 Copies a stringified representation of the source SV into the
3277 destination SV. Automatically performs any necessary C<mg_get> and
3278 coercion of numeric values into strings. Guaranteed to preserve
3279 C<UTF8> flag even from overloaded objects. Similar in nature to
3280 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3281 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3282 would lose the UTF-8'ness of the PV.
3284 =for apidoc sv_copypv_nomg
3286 Like C<sv_copypv>, but doesn't invoke get magic first.
3288 =for apidoc sv_copypv_flags
3290 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3291 has the C<SV_GMAGIC> bit set.
3297 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3302 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3304 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3305 sv_setpvn(dsv,s,len);
3313 =for apidoc sv_2pvbyte
3315 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3316 to its length. May cause the SV to be downgraded from UTF-8 as a
3319 Usually accessed via the C<SvPVbyte> macro.
3325 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3327 PERL_ARGS_ASSERT_SV_2PVBYTE;
3330 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3331 || isGV_with_GP(sv) || SvROK(sv)) {
3332 SV *sv2 = sv_newmortal();
3333 sv_copypv_nomg(sv2,sv);
3336 sv_utf8_downgrade(sv,0);
3337 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3341 =for apidoc sv_2pvutf8
3343 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3344 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3346 Usually accessed via the C<SvPVutf8> macro.
3352 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3354 PERL_ARGS_ASSERT_SV_2PVUTF8;
3356 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3357 || isGV_with_GP(sv) || SvROK(sv))
3358 sv = sv_mortalcopy(sv);
3361 sv_utf8_upgrade_nomg(sv);
3362 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3367 =for apidoc sv_2bool
3369 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3370 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3371 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3373 =for apidoc sv_2bool_flags
3375 This function is only used by C<sv_true()> and friends, and only if
3376 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3377 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3384 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3386 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3389 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3395 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3396 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3399 if(SvGMAGICAL(sv)) {
3401 goto restart; /* call sv_2bool */
3403 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3404 else if(!SvOK(sv)) {
3407 else if(SvPOK(sv)) {
3408 svb = SvPVXtrue(sv);
3410 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3411 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3412 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3416 goto restart; /* call sv_2bool_nomg */
3426 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3428 if (SvNOK(sv) && !SvPOK(sv))
3429 return SvNVX(sv) != 0.0;
3431 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3435 =for apidoc sv_utf8_upgrade
3437 Converts the PV of an SV to its UTF-8-encoded form.
3438 Forces the SV to string form if it is not already.
3439 Will C<mg_get> on C<sv> if appropriate.
3440 Always sets the C<SvUTF8> flag to avoid future validity checks even
3441 if the whole string is the same in UTF-8 as not.
3442 Returns the number of bytes in the converted string
3444 This is not a general purpose byte encoding to Unicode interface:
3445 use the Encode extension for that.
3447 =for apidoc sv_utf8_upgrade_nomg
3449 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3451 =for apidoc sv_utf8_upgrade_flags
3453 Converts the PV of an SV to its UTF-8-encoded form.
3454 Forces the SV to string form if it is not already.
3455 Always sets the SvUTF8 flag to avoid future validity checks even
3456 if all the bytes are invariant in UTF-8.
3457 If C<flags> has C<SV_GMAGIC> bit set,
3458 will C<mg_get> on C<sv> if appropriate, else not.
3460 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3462 Returns the number of bytes in the converted string.
3464 This is not a general purpose byte encoding to Unicode interface:
3465 use the Encode extension for that.
3467 =for apidoc sv_utf8_upgrade_flags_grow
3469 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3470 the number of unused bytes the string of C<sv> is guaranteed to have free after
3471 it upon return. This allows the caller to reserve extra space that it intends
3472 to fill, to avoid extra grows.
3474 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3475 are implemented in terms of this function.
3477 Returns the number of bytes in the converted string (not including the spares).
3481 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3482 C<NUL> isn't guaranteed due to having other routines do the work in some input
3483 cases, or if the input is already flagged as being in utf8.
3488 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3490 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3492 if (sv == &PL_sv_undef)
3494 if (!SvPOK_nog(sv)) {
3496 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3497 (void) sv_2pv_flags(sv,&len, flags);
3499 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3503 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3507 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3508 * compiled and individual nodes will remain non-utf8 even if the
3509 * stringified version of the pattern gets upgraded. Whether the
3510 * PVX of a REGEXP should be grown or we should just croak, I don't
3512 if (SvUTF8(sv) || isREGEXP(sv)) {
3513 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3518 S_sv_uncow(aTHX_ sv, 0);
3521 if (SvCUR(sv) == 0) {
3522 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3524 } else { /* Assume Latin-1/EBCDIC */
3525 /* This function could be much more efficient if we
3526 * had a FLAG in SVs to signal if there are any variant
3527 * chars in the PV. Given that there isn't such a flag
3528 * make the loop as fast as possible. */
3529 U8 * s = (U8 *) SvPVX_const(sv);
3532 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3534 /* utf8 conversion not needed because all are invariants. Mark
3535 * as UTF-8 even if no variant - saves scanning loop */
3537 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3541 /* Here, there is at least one variant (t points to the first one), so
3542 * the string should be converted to utf8. Everything from 's' to
3543 * 't - 1' will occupy only 1 byte each on output.
3545 * Note that the incoming SV may not have a trailing '\0', as certain
3546 * code in pp_formline can send us partially built SVs.
3548 * There are two main ways to convert. One is to create a new string
3549 * and go through the input starting from the beginning, appending each
3550 * converted value onto the new string as we go along. Going this
3551 * route, it's probably best to initially allocate enough space in the
3552 * string rather than possibly running out of space and having to
3553 * reallocate and then copy what we've done so far. Since everything
3554 * from 's' to 't - 1' is invariant, the destination can be initialized
3555 * with these using a fast memory copy. To be sure to allocate enough
3556 * space, one could use the worst case scenario, where every remaining
3557 * byte expands to two under UTF-8, or one could parse it and count
3558 * exactly how many do expand.
3560 * The other way is to unconditionally parse the remainder of the
3561 * string to figure out exactly how big the expanded string will be,
3562 * growing if needed. Then start at the end of the string and place
3563 * the character there at the end of the unfilled space in the expanded
3564 * one, working backwards until reaching 't'.
3566 * The problem with assuming the worst case scenario is that for very
3567 * long strings, we could allocate much more memory than actually
3568 * needed, which can create performance problems. If we have to parse
3569 * anyway, the second method is the winner as it may avoid an extra
3570 * copy. The code used to use the first method under some
3571 * circumstances, but now that there is faster variant counting on
3572 * ASCII platforms, the second method is used exclusively, eliminating
3573 * some code that no longer has to be maintained. */
3576 /* Count the total number of variants there are. We can start
3577 * just beyond the first one, which is known to be at 't' */
3578 const Size_t invariant_length = t - s;
3579 U8 * e = (U8 *) SvEND(sv);
3581 /* The length of the left overs, plus 1. */
3582 const Size_t remaining_length_p1 = e - t;
3584 /* We expand by 1 for the variant at 't' and one for each remaining
3585 * variant (we start looking at 't+1') */
3586 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3588 /* +1 = trailing NUL */
3589 Size_t need = SvCUR(sv) + expansion + extra + 1;
3592 /* Grow if needed */
3593 if (SvLEN(sv) < need) {
3594 t = invariant_length + (U8*) SvGROW(sv, need);
3595 e = t + remaining_length_p1;
3597 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3599 /* Set the NUL at the end */
3600 d = (U8 *) SvEND(sv);
3603 /* Having decremented d, it points to the position to put the
3604 * very last byte of the expanded string. Go backwards through
3605 * the string, copying and expanding as we go, stopping when we
3606 * get to the part that is invariant the rest of the way down */
3610 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3613 *d-- = UTF8_EIGHT_BIT_LO(*e);
3614 *d-- = UTF8_EIGHT_BIT_HI(*e);
3619 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3620 /* Update pos. We do it at the end rather than during
3621 * the upgrade, to avoid slowing down the common case
3622 * (upgrade without pos).
3623 * pos can be stored as either bytes or characters. Since
3624 * this was previously a byte string we can just turn off
3625 * the bytes flag. */
3626 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3628 mg->mg_flags &= ~MGf_BYTES;
3630 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3631 magic_setutf8(sv,mg); /* clear UTF8 cache */
3641 =for apidoc sv_utf8_downgrade
3643 Attempts to convert the PV of an SV from characters to bytes.
3644 If the PV contains a character that cannot fit
3645 in a byte, this conversion will fail;
3646 in this case, either returns false or, if C<fail_ok> is not
3649 This is not a general purpose Unicode to byte encoding interface:
3650 use the C<Encode> extension for that.
3656 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3658 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3660 if (SvPOKp(sv) && SvUTF8(sv)) {
3664 int mg_flags = SV_GMAGIC;
3667 S_sv_uncow(aTHX_ sv, 0);
3669 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3671 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3672 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3673 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3674 SV_GMAGIC|SV_CONST_RETURN);
3675 mg_flags = 0; /* sv_pos_b2u does get magic */
3677 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3678 magic_setutf8(sv,mg); /* clear UTF8 cache */
3681 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3683 if (!utf8_to_bytes(s, &len)) {
3688 Perl_croak(aTHX_ "Wide character in %s",
3691 Perl_croak(aTHX_ "Wide character");
3702 =for apidoc sv_utf8_encode
3704 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3705 flag off so that it looks like octets again.
3711 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3713 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3715 if (SvREADONLY(sv)) {
3716 sv_force_normal_flags(sv, 0);
3718 (void) sv_utf8_upgrade(sv);
3723 =for apidoc sv_utf8_decode
3725 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3726 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3727 so that it looks like a character. If the PV contains only single-byte
3728 characters, the C<SvUTF8> flag stays off.
3729 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3735 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3737 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3740 const U8 *start, *c, *first_variant;
3742 /* The octets may have got themselves encoded - get them back as
3745 if (!sv_utf8_downgrade(sv, TRUE))
3748 /* it is actually just a matter of turning the utf8 flag on, but
3749 * we want to make sure everything inside is valid utf8 first.
3751 c = start = (const U8 *) SvPVX_const(sv);
3752 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3753 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3757 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3758 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3759 after this, clearing pos. Does anything on CPAN
3761 /* adjust pos to the start of a UTF8 char sequence */
3762 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3764 I32 pos = mg->mg_len;
3766 for (c = start + pos; c > start; c--) {
3767 if (UTF8_IS_START(*c))
3770 mg->mg_len = c - start;
3773 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3774 magic_setutf8(sv,mg); /* clear UTF8 cache */
3781 =for apidoc sv_setsv
3783 Copies the contents of the source SV C<ssv> into the destination SV
3784 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3785 function if the source SV needs to be reused. Does not handle 'set' magic on
3786 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3787 performs a copy-by-value, obliterating any previous content of the
3790 You probably want to use one of the assortment of wrappers, such as
3791 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3792 C<SvSetMagicSV_nosteal>.
3794 =for apidoc sv_setsv_flags
3796 Copies the contents of the source SV C<ssv> into the destination SV
3797 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3798 function if the source SV needs to be reused. Does not handle 'set' magic.
3799 Loosely speaking, it performs a copy-by-value, obliterating any previous
3800 content of the destination.
3801 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3802 C<ssv> if appropriate, else not. If the C<flags>
3803 parameter has the C<SV_NOSTEAL> bit set then the
3804 buffers of temps will not be stolen. C<sv_setsv>
3805 and C<sv_setsv_nomg> are implemented in terms of this function.
3807 You probably want to use one of the assortment of wrappers, such as
3808 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3809 C<SvSetMagicSV_nosteal>.
3811 This is the primary function for copying scalars, and most other
3812 copy-ish functions and macros use this underneath.
3818 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3820 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3821 HV *old_stash = NULL;
3823 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3825 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3826 const char * const name = GvNAME(sstr);
3827 const STRLEN len = GvNAMELEN(sstr);
3829 if (dtype >= SVt_PV) {
3835 SvUPGRADE(dstr, SVt_PVGV);
3836 (void)SvOK_off(dstr);
3837 isGV_with_GP_on(dstr);
3839 GvSTASH(dstr) = GvSTASH(sstr);
3841 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3842 gv_name_set(MUTABLE_GV(dstr), name, len,
3843 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3844 SvFAKE_on(dstr); /* can coerce to non-glob */
3847 if(GvGP(MUTABLE_GV(sstr))) {
3848 /* If source has method cache entry, clear it */
3850 SvREFCNT_dec(GvCV(sstr));
3851 GvCV_set(sstr, NULL);
3854 /* If source has a real method, then a method is
3857 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3863 /* If dest already had a real method, that's a change as well */
3865 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3866 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3871 /* We don't need to check the name of the destination if it was not a
3872 glob to begin with. */
3873 if(dtype == SVt_PVGV) {
3874 const char * const name = GvNAME((const GV *)dstr);
3875 const STRLEN len = GvNAMELEN(dstr);
3876 if(memEQs(name, len, "ISA")
3877 /* The stash may have been detached from the symbol table, so
3879 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3883 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3884 || (len == 1 && name[0] == ':')) {
3887 /* Set aside the old stash, so we can reset isa caches on
3889 if((old_stash = GvHV(dstr)))
3890 /* Make sure we do not lose it early. */
3891 SvREFCNT_inc_simple_void_NN(
3892 sv_2mortal((SV *)old_stash)
3897 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3900 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3901 * so temporarily protect it */
3903 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3904 gp_free(MUTABLE_GV(dstr));
3905 GvINTRO_off(dstr); /* one-shot flag */
3906 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3909 if (SvTAINTED(sstr))
3911 if (GvIMPORTED(dstr) != GVf_IMPORTED
3912 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3914 GvIMPORTED_on(dstr);
3917 if(mro_changes == 2) {
3918 if (GvAV((const GV *)sstr)) {
3920 SV * const sref = (SV *)GvAV((const GV *)dstr);
3921 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3922 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3923 AV * const ary = newAV();
3924 av_push(ary, mg->mg_obj); /* takes the refcount */
3925 mg->mg_obj = (SV *)ary;
3927 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3929 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3931 mro_isa_changed_in(GvSTASH(dstr));
3933 else if(mro_changes == 3) {
3934 HV * const stash = GvHV(dstr);
3935 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3941 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3942 if (GvIO(dstr) && dtype == SVt_PVGV) {
3943 DEBUG_o(Perl_deb(aTHX_
3944 "glob_assign_glob clearing PL_stashcache\n"));
3945 /* It's a cache. It will rebuild itself quite happily.
3946 It's a lot of effort to work out exactly which key (or keys)
3947 might be invalidated by the creation of the this file handle.
3949 hv_clear(PL_stashcache);
3955 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3957 SV * const sref = SvRV(sstr);
3959 const int intro = GvINTRO(dstr);
3962 const U32 stype = SvTYPE(sref);
3964 PERL_ARGS_ASSERT_GV_SETREF;
3967 GvINTRO_off(dstr); /* one-shot flag */
3968 GvLINE(dstr) = CopLINE(PL_curcop);
3969 GvEGV(dstr) = MUTABLE_GV(dstr);
3974 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3975 import_flag = GVf_IMPORTED_CV;
3978 location = (SV **) &GvHV(dstr);
3979 import_flag = GVf_IMPORTED_HV;
3982 location = (SV **) &GvAV(dstr);
3983 import_flag = GVf_IMPORTED_AV;
3986 location = (SV **) &GvIOp(dstr);
3989 location = (SV **) &GvFORM(dstr);
3992 location = &GvSV(dstr);
3993 import_flag = GVf_IMPORTED_SV;
3996 if (stype == SVt_PVCV) {
3997 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3998 if (GvCVGEN(dstr)) {
3999 SvREFCNT_dec(GvCV(dstr));
4000 GvCV_set(dstr, NULL);
4001 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4004 /* SAVEt_GVSLOT takes more room on the savestack and has more
4005 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4006 leave_scope needs access to the GV so it can reset method
4007 caches. We must use SAVEt_GVSLOT whenever the type is
4008 SVt_PVCV, even if the stash is anonymous, as the stash may
4009 gain a name somehow before leave_scope. */
4010 if (stype == SVt_PVCV) {
4011 /* There is no save_pushptrptrptr. Creating it for this
4012 one call site would be overkill. So inline the ss add
4016 SS_ADD_PTR(location);
4017 SS_ADD_PTR(SvREFCNT_inc(*location));
4018 SS_ADD_UV(SAVEt_GVSLOT);
4021 else SAVEGENERICSV(*location);
4024 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4025 CV* const cv = MUTABLE_CV(*location);
4027 if (!GvCVGEN((const GV *)dstr) &&
4028 (CvROOT(cv) || CvXSUB(cv)) &&
4029 /* redundant check that avoids creating the extra SV
4030 most of the time: */
4031 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4033 SV * const new_const_sv =
4034 CvCONST((const CV *)sref)
4035 ? cv_const_sv((const CV *)sref)
4037 HV * const stash = GvSTASH((const GV *)dstr);
4038 report_redefined_cv(
4041 ? Perl_newSVpvf(aTHX_
4042 "%" HEKf "::%" HEKf,
4043 HEKfARG(HvNAME_HEK(stash)),
4044 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4045 : Perl_newSVpvf(aTHX_
4047 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4050 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4054 cv_ckproto_len_flags(cv, (const GV *)dstr,
4055 SvPOK(sref) ? CvPROTO(sref) : NULL,
4056 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4057 SvPOK(sref) ? SvUTF8(sref) : 0);
4059 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4060 GvASSUMECV_on(dstr);
4061 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4062 if (intro && GvREFCNT(dstr) > 1) {
4063 /* temporary remove extra savestack's ref */
4065 gv_method_changed(dstr);
4068 else gv_method_changed(dstr);
4071 *location = SvREFCNT_inc_simple_NN(sref);
4072 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4073 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4074 GvFLAGS(dstr) |= import_flag;
4077 if (stype == SVt_PVHV) {
4078 const char * const name = GvNAME((GV*)dstr);
4079 const STRLEN len = GvNAMELEN(dstr);
4082 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4083 || (len == 1 && name[0] == ':')
4085 && (!dref || HvENAME_get(dref))
4088 (HV *)sref, (HV *)dref,
4094 stype == SVt_PVAV && sref != dref
4095 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4096 /* The stash may have been detached from the symbol table, so
4097 check its name before doing anything. */
4098 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4101 MAGIC * const omg = dref && SvSMAGICAL(dref)
4102 ? mg_find(dref, PERL_MAGIC_isa)
4104 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4105 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4106 AV * const ary = newAV();
4107 av_push(ary, mg->mg_obj); /* takes the refcount */
4108 mg->mg_obj = (SV *)ary;
4111 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4112 SV **svp = AvARRAY((AV *)omg->mg_obj);
4113 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4117 SvREFCNT_inc_simple_NN(*svp++)
4123 SvREFCNT_inc_simple_NN(omg->mg_obj)
4127 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4133 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4135 for (i = 0; i <= AvFILL(sref); ++i) {
4136 SV **elem = av_fetch ((AV*)sref, i, 0);
4139 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4143 mg = mg_find(sref, PERL_MAGIC_isa);
4145 /* Since the *ISA assignment could have affected more than
4146 one stash, don't call mro_isa_changed_in directly, but let
4147 magic_clearisa do it for us, as it already has the logic for
4148 dealing with globs vs arrays of globs. */
4150 Perl_magic_clearisa(aTHX_ NULL, mg);
4152 else if (stype == SVt_PVIO) {
4153 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4154 /* It's a cache. It will rebuild itself quite happily.
4155 It's a lot of effort to work out exactly which key (or keys)
4156 might be invalidated by the creation of the this file handle.
4158 hv_clear(PL_stashcache);
4162 if (!intro) SvREFCNT_dec(dref);
4163 if (SvTAINTED(sstr))
4171 #ifdef PERL_DEBUG_READONLY_COW
4172 # include <sys/mman.h>
4174 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4175 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4179 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4181 struct perl_memory_debug_header * const header =
4182 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4183 const MEM_SIZE len = header->size;
4184 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4185 # ifdef PERL_TRACK_MEMPOOL
4186 if (!header->readonly) header->readonly = 1;
4188 if (mprotect(header, len, PROT_READ))
4189 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4190 header, len, errno);
4194 S_sv_buf_to_rw(pTHX_ SV *sv)
4196 struct perl_memory_debug_header * const header =
4197 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4198 const MEM_SIZE len = header->size;
4199 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4200 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4201 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4202 header, len, errno);
4203 # ifdef PERL_TRACK_MEMPOOL
4204 header->readonly = 0;
4209 # define sv_buf_to_ro(sv) NOOP
4210 # define sv_buf_to_rw(sv) NOOP
4214 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4219 unsigned int both_type;
4221 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4223 if (UNLIKELY( sstr == dstr ))
4226 if (UNLIKELY( !sstr ))
4227 sstr = &PL_sv_undef;
4229 stype = SvTYPE(sstr);
4230 dtype = SvTYPE(dstr);
4231 both_type = (stype | dtype);
4233 /* with these values, we can check that both SVs are NULL/IV (and not
4234 * freed) just by testing the or'ed types */
4235 STATIC_ASSERT_STMT(SVt_NULL == 0);
4236 STATIC_ASSERT_STMT(SVt_IV == 1);
4237 if (both_type <= 1) {
4238 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4244 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4245 if (SvREADONLY(dstr))
4246 Perl_croak_no_modify();
4248 if (SvWEAKREF(dstr))
4249 sv_unref_flags(dstr, 0);
4251 old_rv = SvRV(dstr);
4254 assert(!SvGMAGICAL(sstr));
4255 assert(!SvGMAGICAL(dstr));
4257 sflags = SvFLAGS(sstr);
4258 if (sflags & (SVf_IOK|SVf_ROK)) {
4259 SET_SVANY_FOR_BODYLESS_IV(dstr);
4260 new_dflags = SVt_IV;
4262 if (sflags & SVf_ROK) {
4263 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4264 new_dflags |= SVf_ROK;
4267 /* both src and dst are <= SVt_IV, so sv_any points to the
4268 * head; so access the head directly
4270 assert( &(sstr->sv_u.svu_iv)
4271 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4272 assert( &(dstr->sv_u.svu_iv)
4273 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4274 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4275 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4279 new_dflags = dtype; /* turn off everything except the type */
4281 SvFLAGS(dstr) = new_dflags;
4282 SvREFCNT_dec(old_rv);
4287 if (UNLIKELY(both_type == SVTYPEMASK)) {
4288 if (SvIS_FREED(dstr)) {
4289 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4290 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4292 if (SvIS_FREED(sstr)) {
4293 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4294 (void*)sstr, (void*)dstr);
4300 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4301 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4303 /* There's a lot of redundancy below but we're going for speed here */
4308 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4309 (void)SvOK_off(dstr);
4317 /* For performance, we inline promoting to type SVt_IV. */
4318 /* We're starting from SVt_NULL, so provided that define is
4319 * actual 0, we don't have to unset any SV type flags
4320 * to promote to SVt_IV. */
4321 STATIC_ASSERT_STMT(SVt_NULL == 0);
4322 SET_SVANY_FOR_BODYLESS_IV(dstr);
4323 SvFLAGS(dstr) |= SVt_IV;
4327 sv_upgrade(dstr, SVt_PVIV);
4331 goto end_of_first_switch;
4333 (void)SvIOK_only(dstr);
4334 SvIV_set(dstr, SvIVX(sstr));
4337 /* SvTAINTED can only be true if the SV has taint magic, which in
4338 turn means that the SV type is PVMG (or greater). This is the
4339 case statement for SVt_IV, so this cannot be true (whatever gcov
4341 assert(!SvTAINTED(sstr));
4346 if (dtype < SVt_PV && dtype != SVt_IV)
4347 sv_upgrade(dstr, SVt_IV);
4351 if (LIKELY( SvNOK(sstr) )) {
4355 sv_upgrade(dstr, SVt_NV);
4359 sv_upgrade(dstr, SVt_PVNV);
4363 goto end_of_first_switch;
4365 SvNV_set(dstr, SvNVX(sstr));
4366 (void)SvNOK_only(dstr);
4367 /* SvTAINTED can only be true if the SV has taint magic, which in
4368 turn means that the SV type is PVMG (or greater). This is the
4369 case statement for SVt_NV, so this cannot be true (whatever gcov
4371 assert(!SvTAINTED(sstr));
4378 sv_upgrade(dstr, SVt_PV);
4381 if (dtype < SVt_PVIV)
4382 sv_upgrade(dstr, SVt_PVIV);
4385 if (dtype < SVt_PVNV)
4386 sv_upgrade(dstr, SVt_PVNV);
4390 invlist_clone(sstr, dstr);
4394 const char * const type = sv_reftype(sstr,0);
4396 /* diag_listed_as: Bizarre copy of %s */
4397 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4399 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4401 NOT_REACHED; /* NOTREACHED */
4405 if (dtype < SVt_REGEXP)
4406 sv_upgrade(dstr, SVt_REGEXP);
4412 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4414 if (SvTYPE(sstr) != stype)
4415 stype = SvTYPE(sstr);
4417 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4418 glob_assign_glob(dstr, sstr, dtype);
4421 if (stype == SVt_PVLV)
4423 if (isREGEXP(sstr)) goto upgregexp;
4424 SvUPGRADE(dstr, SVt_PVNV);
4427 SvUPGRADE(dstr, (svtype)stype);
4429 end_of_first_switch:
4431 /* dstr may have been upgraded. */
4432 dtype = SvTYPE(dstr);
4433 sflags = SvFLAGS(sstr);
4435 if (UNLIKELY( dtype == SVt_PVCV )) {
4436 /* Assigning to a subroutine sets the prototype. */
4439 const char *const ptr = SvPV_const(sstr, len);
4441 SvGROW(dstr, len + 1);
4442 Copy(ptr, SvPVX(dstr), len + 1, char);
4443 SvCUR_set(dstr, len);
4445 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4446 CvAUTOLOAD_off(dstr);
4451 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4452 || dtype == SVt_PVFM))
4454 const char * const type = sv_reftype(dstr,0);
4456 /* diag_listed_as: Cannot copy to %s */
4457 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4459 Perl_croak(aTHX_ "Cannot copy to %s", type);
4460 } else if (sflags & SVf_ROK) {
4461 if (isGV_with_GP(dstr)
4462 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4465 if (GvIMPORTED(dstr) != GVf_IMPORTED
4466 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4468 GvIMPORTED_on(dstr);
4473 glob_assign_glob(dstr, sstr, dtype);
4477 if (dtype >= SVt_PV) {
4478 if (isGV_with_GP(dstr)) {
4479 gv_setref(dstr, sstr);
4482 if (SvPVX_const(dstr)) {
4488 (void)SvOK_off(dstr);
4489 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4490 SvFLAGS(dstr) |= sflags & SVf_ROK;
4491 assert(!(sflags & SVp_NOK));
4492 assert(!(sflags & SVp_IOK));
4493 assert(!(sflags & SVf_NOK));
4494 assert(!(sflags & SVf_IOK));
4496 else if (isGV_with_GP(dstr)) {
4497 if (!(sflags & SVf_OK)) {
4498 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4499 "Undefined value assigned to typeglob");
4502 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4503 if (dstr != (const SV *)gv) {
4504 const char * const name = GvNAME((const GV *)dstr);
4505 const STRLEN len = GvNAMELEN(dstr);
4506 HV *old_stash = NULL;
4507 bool reset_isa = FALSE;
4508 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4509 || (len == 1 && name[0] == ':')) {
4510 /* Set aside the old stash, so we can reset isa caches
4511 on its subclasses. */
4512 if((old_stash = GvHV(dstr))) {
4513 /* Make sure we do not lose it early. */
4514 SvREFCNT_inc_simple_void_NN(
4515 sv_2mortal((SV *)old_stash)
4522 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4523 gp_free(MUTABLE_GV(dstr));
4525 GvGP_set(dstr, gp_ref(GvGP(gv)));
4528 HV * const stash = GvHV(dstr);
4530 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4540 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4541 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4542 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4544 else if (sflags & SVp_POK) {
4545 const STRLEN cur = SvCUR(sstr);
4546 const STRLEN len = SvLEN(sstr);
4549 * We have three basic ways to copy the string:
4555 * Which we choose is based on various factors. The following
4556 * things are listed in order of speed, fastest to slowest:
4558 * - Copying a short string
4559 * - Copy-on-write bookkeeping
4561 * - Copying a long string
4563 * We swipe the string (steal the string buffer) if the SV on the
4564 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4565 * big win on long strings. It should be a win on short strings if
4566 * SvPVX_const(dstr) has to be allocated. If not, it should not
4567 * slow things down, as SvPVX_const(sstr) would have been freed
4570 * We also steal the buffer from a PADTMP (operator target) if it
4571 * is ‘long enough’. For short strings, a swipe does not help
4572 * here, as it causes more malloc calls the next time the target
4573 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4574 * be allocated it is still not worth swiping PADTMPs for short
4575 * strings, as the savings here are small.
4577 * If swiping is not an option, then we see whether it is
4578 * worth using copy-on-write. If the lhs already has a buf-
4579 * fer big enough and the string is short, we skip it and fall back
4580 * to method 3, since memcpy is faster for short strings than the
4581 * later bookkeeping overhead that copy-on-write entails.
4583 * If the rhs is not a copy-on-write string yet, then we also
4584 * consider whether the buffer is too large relative to the string
4585 * it holds. Some operations such as readline allocate a large
4586 * buffer in the expectation of reusing it. But turning such into
4587 * a COW buffer is counter-productive because it increases memory
4588 * usage by making readline allocate a new large buffer the sec-
4589 * ond time round. So, if the buffer is too large, again, we use
4592 * Finally, if there is no buffer on the left, or the buffer is too
4593 * small, then we use copy-on-write and make both SVs share the
4598 /* Whichever path we take through the next code, we want this true,
4599 and doing it now facilitates the COW check. */
4600 (void)SvPOK_only(dstr);
4604 /* slated for free anyway (and not COW)? */
4605 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4606 /* or a swipable TARG */
4608 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4610 /* whose buffer is worth stealing */
4611 && CHECK_COWBUF_THRESHOLD(cur,len)
4614 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4615 (!(flags & SV_NOSTEAL)) &&
4616 /* and we're allowed to steal temps */
4617 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4618 len) /* and really is a string */
4619 { /* Passes the swipe test. */
4620 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4622 SvPV_set(dstr, SvPVX_mutable(sstr));
4623 SvLEN_set(dstr, SvLEN(sstr));
4624 SvCUR_set(dstr, SvCUR(sstr));
4627 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4628 SvPV_set(sstr, NULL);
4633 else if (flags & SV_COW_SHARED_HASH_KEYS
4635 #ifdef PERL_COPY_ON_WRITE
4638 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4639 /* If this is a regular (non-hek) COW, only so
4640 many COW "copies" are possible. */
4641 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4642 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4643 && !(SvFLAGS(dstr) & SVf_BREAK)
4644 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4645 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4649 && !(SvFLAGS(dstr) & SVf_BREAK)
4652 /* Either it's a shared hash key, or it's suitable for
4656 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4662 if (!(sflags & SVf_IsCOW)) {
4664 CowREFCNT(sstr) = 0;
4667 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4673 if (sflags & SVf_IsCOW) {
4677 SvPV_set(dstr, SvPVX_mutable(sstr));
4682 /* SvIsCOW_shared_hash */
4683 DEBUG_C(PerlIO_printf(Perl_debug_log,
4684 "Copy on write: Sharing hash\n"));
4686 assert (SvTYPE(dstr) >= SVt_PV);
4688 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4690 SvLEN_set(dstr, len);
4691 SvCUR_set(dstr, cur);
4694 /* Failed the swipe test, and we cannot do copy-on-write either.
4695 Have to copy the string. */
4696 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4697 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4698 SvCUR_set(dstr, cur);
4699 *SvEND(dstr) = '\0';
4701 if (sflags & SVp_NOK) {
4702 SvNV_set(dstr, SvNVX(sstr));
4704 if (sflags & SVp_IOK) {
4705 SvIV_set(dstr, SvIVX(sstr));
4706 if (sflags & SVf_IVisUV)
4709 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4711 const MAGIC * const smg = SvVSTRING_mg(sstr);
4713 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4714 smg->mg_ptr, smg->mg_len);
4715 SvRMAGICAL_on(dstr);
4719 else if (sflags & (SVp_IOK|SVp_NOK)) {
4720 (void)SvOK_off(dstr);
4721 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4722 if (sflags & SVp_IOK) {
4723 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4724 SvIV_set(dstr, SvIVX(sstr));
4726 if (sflags & SVp_NOK) {
4727 SvNV_set(dstr, SvNVX(sstr));
4731 if (isGV_with_GP(sstr)) {
4732 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4735 (void)SvOK_off(dstr);
4737 if (SvTAINTED(sstr))
4743 =for apidoc sv_set_undef
4745 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4746 Doesn't handle set magic.
4748 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4749 buffer, unlike C<undef $sv>.
4751 Introduced in perl 5.25.12.
4757 Perl_sv_set_undef(pTHX_ SV *sv)
4759 U32 type = SvTYPE(sv);
4761 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4763 /* shortcut, NULL, IV, RV */
4765 if (type <= SVt_IV) {
4766 assert(!SvGMAGICAL(sv));
4767 if (SvREADONLY(sv)) {
4768 /* does undeffing PL_sv_undef count as modifying a read-only
4769 * variable? Some XS code does this */
4770 if (sv == &PL_sv_undef)
4772 Perl_croak_no_modify();
4777 sv_unref_flags(sv, 0);
4780 SvFLAGS(sv) = type; /* quickly turn off all flags */
4781 SvREFCNT_dec_NN(rv);
4785 SvFLAGS(sv) = type; /* quickly turn off all flags */
4790 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4793 SV_CHECK_THINKFIRST_COW_DROP(sv);
4795 if (isGV_with_GP(sv))
4796 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4797 "Undefined value assigned to typeglob");
4805 =for apidoc sv_setsv_mg
4807 Like C<sv_setsv>, but also handles 'set' magic.
4813 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4815 PERL_ARGS_ASSERT_SV_SETSV_MG;
4817 sv_setsv(dstr,sstr);
4822 # define SVt_COW SVt_PV
4824 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4826 STRLEN cur = SvCUR(sstr);
4827 STRLEN len = SvLEN(sstr);
4829 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4830 const bool already = cBOOL(SvIsCOW(sstr));
4833 PERL_ARGS_ASSERT_SV_SETSV_COW;
4836 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4837 (void*)sstr, (void*)dstr);
4844 if (SvTHINKFIRST(dstr))
4845 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4846 else if (SvPVX_const(dstr))
4847 Safefree(SvPVX_mutable(dstr));
4851 SvUPGRADE(dstr, SVt_COW);
4853 assert (SvPOK(sstr));
4854 assert (SvPOKp(sstr));
4856 if (SvIsCOW(sstr)) {
4858 if (SvLEN(sstr) == 0) {
4859 /* source is a COW shared hash key. */
4860 DEBUG_C(PerlIO_printf(Perl_debug_log,
4861 "Fast copy on write: Sharing hash\n"));
4862 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4865 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4866 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4868 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4869 SvUPGRADE(sstr, SVt_COW);
4871 DEBUG_C(PerlIO_printf(Perl_debug_log,
4872 "Fast copy on write: Converting sstr to COW\n"));
4873 CowREFCNT(sstr) = 0;
4875 # ifdef PERL_DEBUG_READONLY_COW
4876 if (already) sv_buf_to_rw(sstr);
4879 new_pv = SvPVX_mutable(sstr);
4883 SvPV_set(dstr, new_pv);
4884 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4887 SvLEN_set(dstr, len);
4888 SvCUR_set(dstr, cur);
4898 =for apidoc sv_setpv_bufsize
4900 Sets the SV to be a string of cur bytes length, with at least
4901 len bytes available. Ensures that there is a null byte at SvEND.
4902 Returns a char * pointer to the SvPV buffer.
4908 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4912 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4914 SV_CHECK_THINKFIRST_COW_DROP(sv);
4915 SvUPGRADE(sv, SVt_PV);
4916 pv = SvGROW(sv, len + 1);
4919 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4922 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4927 =for apidoc sv_setpvn
4929 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4930 The C<len> parameter indicates the number of
4931 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4932 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4938 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4942 PERL_ARGS_ASSERT_SV_SETPVN;
4944 SV_CHECK_THINKFIRST_COW_DROP(sv);
4945 if (isGV_with_GP(sv))
4946 Perl_croak_no_modify();
4952 /* len is STRLEN which is unsigned, need to copy to signed */
4955 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4958 SvUPGRADE(sv, SVt_PV);
4960 dptr = SvGROW(sv, len + 1);
4961 Move(ptr,dptr,len,char);
4964 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4966 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4970 =for apidoc sv_setpvn_mg
4972 Like C<sv_setpvn>, but also handles 'set' magic.
4978 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4980 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4982 sv_setpvn(sv,ptr,len);
4987 =for apidoc sv_setpv
4989 Copies a string into an SV. The string must be terminated with a C<NUL>
4990 character, and not contain embeded C<NUL>'s.
4991 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4997 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5001 PERL_ARGS_ASSERT_SV_SETPV;
5003 SV_CHECK_THINKFIRST_COW_DROP(sv);
5009 SvUPGRADE(sv, SVt_PV);
5011 SvGROW(sv, len + 1);
5012 Move(ptr,SvPVX(sv),len+1,char);
5014 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5016 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5020 =for apidoc sv_setpv_mg
5022 Like C<sv_setpv>, but also handles 'set' magic.
5028 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5030 PERL_ARGS_ASSERT_SV_SETPV_MG;
5037 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5039 PERL_ARGS_ASSERT_SV_SETHEK;
5045 if (HEK_LEN(hek) == HEf_SVKEY) {
5046 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5049 const int flags = HEK_FLAGS(hek);
5050 if (flags & HVhek_WASUTF8) {
5051 STRLEN utf8_len = HEK_LEN(hek);
5052 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5053 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5056 } else if (flags & HVhek_UNSHARED) {
5057 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5060 else SvUTF8_off(sv);
5064 SV_CHECK_THINKFIRST_COW_DROP(sv);
5065 SvUPGRADE(sv, SVt_PV);
5067 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5068 SvCUR_set(sv, HEK_LEN(hek));
5074 else SvUTF8_off(sv);
5082 =for apidoc sv_usepvn_flags
5084 Tells an SV to use C<ptr> to find its string value. Normally the
5085 string is stored inside the SV, but sv_usepvn allows the SV to use an
5086 outside string. C<ptr> should point to memory that was allocated
5087 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5088 the start of a C<Newx>-ed block of memory, and not a pointer to the
5089 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5090 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5091 string length, C<len>, must be supplied. By default this function
5092 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5093 so that pointer should not be freed or used by the programmer after
5094 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5095 that pointer (e.g. ptr + 1) be used.
5097 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5098 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5100 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5101 C<len>, and already meets the requirements for storing in C<SvPVX>).
5107 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5111 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5113 SV_CHECK_THINKFIRST_COW_DROP(sv);
5114 SvUPGRADE(sv, SVt_PV);
5117 if (flags & SV_SMAGIC)
5121 if (SvPVX_const(sv))
5125 if (flags & SV_HAS_TRAILING_NUL)
5126 assert(ptr[len] == '\0');
5129 allocate = (flags & SV_HAS_TRAILING_NUL)
5131 #ifdef Perl_safesysmalloc_size
5134 PERL_STRLEN_ROUNDUP(len + 1);
5136 if (flags & SV_HAS_TRAILING_NUL) {
5137 /* It's long enough - do nothing.
5138 Specifically Perl_newCONSTSUB is relying on this. */
5141 /* Force a move to shake out bugs in callers. */
5142 char *new_ptr = (char*)safemalloc(allocate);
5143 Copy(ptr, new_ptr, len, char);
5144 PoisonFree(ptr,len,char);
5148 ptr = (char*) saferealloc (ptr, allocate);
5151 #ifdef Perl_safesysmalloc_size
5152 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5154 SvLEN_set(sv, allocate);
5158 if (!(flags & SV_HAS_TRAILING_NUL)) {
5161 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5163 if (flags & SV_SMAGIC)
5169 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5171 assert(SvIsCOW(sv));
5174 const char * const pvx = SvPVX_const(sv);
5175 const STRLEN len = SvLEN(sv);
5176 const STRLEN cur = SvCUR(sv);
5180 PerlIO_printf(Perl_debug_log,
5181 "Copy on write: Force normal %ld\n",
5187 # ifdef PERL_COPY_ON_WRITE
5189 /* Must do this first, since the CowREFCNT uses SvPVX and
5190 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5191 the only owner left of the buffer. */
5192 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5194 U8 cowrefcnt = CowREFCNT(sv);
5195 if(cowrefcnt != 0) {
5197 CowREFCNT(sv) = cowrefcnt;
5202 /* Else we are the only owner of the buffer. */
5207 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5212 if (flags & SV_COW_DROP_PV) {
5213 /* OK, so we don't need to copy our buffer. */
5216 SvGROW(sv, cur + 1);
5217 Move(pvx,SvPVX(sv),cur,char);
5222 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5230 const char * const pvx = SvPVX_const(sv);
5231 const STRLEN len = SvCUR(sv);
5235 if (flags & SV_COW_DROP_PV) {
5236 /* OK, so we don't need to copy our buffer. */
5239 SvGROW(sv, len + 1);
5240 Move(pvx,SvPVX(sv),len,char);
5243 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5250 =for apidoc sv_force_normal_flags
5252 Undo various types of fakery on an SV, where fakery means
5253 "more than" a string: if the PV is a shared string, make
5254 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5255 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5256 we do the copy, and is also used locally; if this is a
5257 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5258 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5259 C<SvPOK_off> rather than making a copy. (Used where this
5260 scalar is about to be set to some other value.) In addition,
5261 the C<flags> parameter gets passed to C<sv_unref_flags()>
5262 when unreffing. C<sv_force_normal> calls this function
5263 with flags set to 0.
5265 This function is expected to be used to signal to perl that this SV is
5266 about to be written to, and any extra book-keeping needs to be taken care
5267 of. Hence, it croaks on read-only values.
5273 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5275 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5278 Perl_croak_no_modify();
5279 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5280 S_sv_uncow(aTHX_ sv, flags);
5282 sv_unref_flags(sv, flags);
5283 else if (SvFAKE(sv) && isGV_with_GP(sv))
5284 sv_unglob(sv, flags);
5285 else if (SvFAKE(sv) && isREGEXP(sv)) {
5286 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5287 to sv_unglob. We only need it here, so inline it. */
5288 const bool islv = SvTYPE(sv) == SVt_PVLV;
5289 const svtype new_type =
5290 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5291 SV *const temp = newSV_type(new_type);
5292 regexp *old_rx_body;
5294 if (new_type == SVt_PVMG) {
5295 SvMAGIC_set(temp, SvMAGIC(sv));
5296 SvMAGIC_set(sv, NULL);
5297 SvSTASH_set(temp, SvSTASH(sv));
5298 SvSTASH_set(sv, NULL);
5301 SvCUR_set(temp, SvCUR(sv));
5302 /* Remember that SvPVX is in the head, not the body. */
5303 assert(ReANY((REGEXP *)sv)->mother_re);
5306 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5307 * whose xpvlenu_rx field points to the regex body */
5308 XPV *xpv = (XPV*)(SvANY(sv));
5309 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5310 xpv->xpv_len_u.xpvlenu_rx = NULL;
5313 old_rx_body = ReANY((REGEXP *)sv);
5315 /* Their buffer is already owned by someone else. */
5316 if (flags & SV_COW_DROP_PV) {
5317 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5318 zeroed body. For SVt_PVLV, we zeroed it above (len field
5319 a union with xpvlenu_rx) */
5320 assert(!SvLEN(islv ? sv : temp));
5321 sv->sv_u.svu_pv = 0;
5324 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5325 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5329 /* Now swap the rest of the bodies. */
5333 SvFLAGS(sv) &= ~SVTYPEMASK;
5334 SvFLAGS(sv) |= new_type;
5335 SvANY(sv) = SvANY(temp);
5338 SvFLAGS(temp) &= ~(SVTYPEMASK);
5339 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5340 SvANY(temp) = old_rx_body;
5342 SvREFCNT_dec_NN(temp);
5344 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5350 Efficient removal of characters from the beginning of the string buffer.
5351 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5352 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5353 character of the adjusted string. Uses the C<OOK> hack. On return, only
5354 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5356 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5357 refer to the same chunk of data.
5359 The unfortunate similarity of this function's name to that of Perl's C<chop>
5360 operator is strictly coincidental. This function works from the left;
5361 C<chop> works from the right.
5367 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5378 PERL_ARGS_ASSERT_SV_CHOP;
5380 if (!ptr || !SvPOKp(sv))
5382 delta = ptr - SvPVX_const(sv);
5384 /* Nothing to do. */
5387 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5388 if (delta > max_delta)
5389 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5390 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5391 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5392 SV_CHECK_THINKFIRST(sv);
5393 SvPOK_only_UTF8(sv);
5396 if (!SvLEN(sv)) { /* make copy of shared string */
5397 const char *pvx = SvPVX_const(sv);
5398 const STRLEN len = SvCUR(sv);
5399 SvGROW(sv, len + 1);
5400 Move(pvx,SvPVX(sv),len,char);
5406 SvOOK_offset(sv, old_delta);
5408 SvLEN_set(sv, SvLEN(sv) - delta);
5409 SvCUR_set(sv, SvCUR(sv) - delta);
5410 SvPV_set(sv, SvPVX(sv) + delta);
5412 p = (U8 *)SvPVX_const(sv);
5415 /* how many bytes were evacuated? we will fill them with sentinel
5416 bytes, except for the part holding the new offset of course. */
5419 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5421 assert(evacn <= delta + old_delta);
5425 /* This sets 'delta' to the accumulated value of all deltas so far */
5429 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5430 * the string; otherwise store a 0 byte there and store 'delta' just prior
5431 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5432 * portion of the chopped part of the string */
5433 if (delta < 0x100) {
5437 p -= sizeof(STRLEN);
5438 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5442 /* Fill the preceding buffer with sentinals to verify that no-one is
5452 =for apidoc sv_catpvn
5454 Concatenates the string onto the end of the string which is in the SV.
5455 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5456 status set, then the bytes appended should be valid UTF-8.
5457 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5459 =for apidoc sv_catpvn_flags
5461 Concatenates the string onto the end of the string which is in the SV. The
5462 C<len> indicates number of bytes to copy.
5464 By default, the string appended is assumed to be valid UTF-8 if the SV has
5465 the UTF-8 status set, and a string of bytes otherwise. One can force the
5466 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5467 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5468 string appended will be upgraded to UTF-8 if necessary.
5470 If C<flags> has the C<SV_SMAGIC> bit set, will
5471 C<mg_set> on C<dsv> afterwards if appropriate.
5472 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5473 in terms of this function.
5479 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5482 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5484 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5485 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5487 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5488 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5489 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5492 else SvGROW(dsv, dlen + slen + 3);
5494 sstr = SvPVX_const(dsv);
5495 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5496 SvCUR_set(dsv, SvCUR(dsv) + slen);
5499 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5500 const char * const send = sstr + slen;
5503 /* Something this code does not account for, which I think is
5504 impossible; it would require the same pv to be treated as
5505 bytes *and* utf8, which would indicate a bug elsewhere. */
5506 assert(sstr != dstr);
5508 SvGROW(dsv, dlen + slen * 2 + 3);
5509 d = (U8 *)SvPVX(dsv) + dlen;
5511 while (sstr < send) {
5512 append_utf8_from_native_byte(*sstr, &d);
5515 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5518 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5520 if (flags & SV_SMAGIC)
5525 =for apidoc sv_catsv
5527 Concatenates the string from SV C<ssv> onto the end of the string in SV
5528 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5529 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5530 and C<L</sv_catsv_nomg>>.
5532 =for apidoc sv_catsv_flags
5534 Concatenates the string from SV C<ssv> onto the end of the string in SV
5535 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5536 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5537 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5538 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5539 and C<sv_catsv_mg> are implemented in terms of this function.
5544 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5546 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5550 const char *spv = SvPV_flags_const(ssv, slen, flags);
5551 if (flags & SV_GMAGIC)
5553 sv_catpvn_flags(dsv, spv, slen,
5554 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5555 if (flags & SV_SMAGIC)
5561 =for apidoc sv_catpv
5563 Concatenates the C<NUL>-terminated string onto the end of the string which is
5565 If the SV has the UTF-8 status set, then the bytes appended should be
5566 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5572 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5578 PERL_ARGS_ASSERT_SV_CATPV;
5582 junk = SvPV_force(sv, tlen);
5584 SvGROW(sv, tlen + len + 1);
5586 ptr = SvPVX_const(sv);
5587 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5588 SvCUR_set(sv, SvCUR(sv) + len);
5589 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5594 =for apidoc sv_catpv_flags
5596 Concatenates the C<NUL>-terminated string onto the end of the string which is
5598 If the SV has the UTF-8 status set, then the bytes appended should
5599 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5600 on the modified SV if appropriate.
5606 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5608 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5609 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5613 =for apidoc sv_catpv_mg
5615 Like C<sv_catpv>, but also handles 'set' magic.
5621 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5623 PERL_ARGS_ASSERT_SV_CATPV_MG;
5632 Creates a new SV. A non-zero C<len> parameter indicates the number of
5633 bytes of preallocated string space the SV should have. An extra byte for a
5634 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5635 space is allocated.) The reference count for the new SV is set to 1.
5637 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5638 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5639 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5640 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5641 modules supporting older perls.
5647 Perl_newSV(pTHX_ const STRLEN len)
5653 sv_grow(sv, len + 1);
5658 =for apidoc sv_magicext
5660 Adds magic to an SV, upgrading it if necessary. Applies the
5661 supplied C<vtable> and returns a pointer to the magic added.
5663 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5664 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5665 one instance of the same C<how>.
5667 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5668 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5669 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5670 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5672 (This is now used as a subroutine by C<sv_magic>.)
5677 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5678 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5682 PERL_ARGS_ASSERT_SV_MAGICEXT;
5684 SvUPGRADE(sv, SVt_PVMG);
5685 Newxz(mg, 1, MAGIC);
5686 mg->mg_moremagic = SvMAGIC(sv);
5687 SvMAGIC_set(sv, mg);
5689 /* Sometimes a magic contains a reference loop, where the sv and
5690 object refer to each other. To prevent a reference loop that
5691 would prevent such objects being freed, we look for such loops
5692 and if we find one we avoid incrementing the object refcount.
5694 Note we cannot do this to avoid self-tie loops as intervening RV must
5695 have its REFCNT incremented to keep it in existence.
5698 if (!obj || obj == sv ||
5699 how == PERL_MAGIC_arylen ||
5700 how == PERL_MAGIC_regdata ||
5701 how == PERL_MAGIC_regdatum ||
5702 how == PERL_MAGIC_symtab ||
5703 (SvTYPE(obj) == SVt_PVGV &&
5704 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5705 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5706 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5711 mg->mg_obj = SvREFCNT_inc_simple(obj);
5712 mg->mg_flags |= MGf_REFCOUNTED;
5715 /* Normal self-ties simply pass a null object, and instead of
5716 using mg_obj directly, use the SvTIED_obj macro to produce a
5717 new RV as needed. For glob "self-ties", we are tieing the PVIO
5718 with an RV obj pointing to the glob containing the PVIO. In
5719 this case, to avoid a reference loop, we need to weaken the
5723 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5724 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5730 mg->mg_len = namlen;
5733 mg->mg_ptr = savepvn(name, namlen);
5734 else if (namlen == HEf_SVKEY) {
5735 /* Yes, this is casting away const. This is only for the case of
5736 HEf_SVKEY. I think we need to document this aberation of the
5737 constness of the API, rather than making name non-const, as
5738 that change propagating outwards a long way. */
5739 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5741 mg->mg_ptr = (char *) name;
5743 mg->mg_virtual = (MGVTBL *) vtable;
5750 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5752 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5753 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5754 /* This sv is only a delegate. //g magic must be attached to
5759 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5760 &PL_vtbl_mglob, 0, 0);
5764 =for apidoc sv_magic
5766 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5767 necessary, then adds a new magic item of type C<how> to the head of the
5770 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5771 handling of the C<name> and C<namlen> arguments.
5773 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5774 to add more than one instance of the same C<how>.
5780 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5781 const char *const name, const I32 namlen)
5783 const MGVTBL *vtable;
5786 unsigned int vtable_index;
5788 PERL_ARGS_ASSERT_SV_MAGIC;
5790 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5791 || ((flags = PL_magic_data[how]),
5792 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5793 > magic_vtable_max))
5794 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5796 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5797 Useful for attaching extension internal data to perl vars.
5798 Note that multiple extensions may clash if magical scalars
5799 etc holding private data from one are passed to another. */
5801 vtable = (vtable_index == magic_vtable_max)
5802 ? NULL : PL_magic_vtables + vtable_index;
5804 if (SvREADONLY(sv)) {
5806 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5809 Perl_croak_no_modify();
5812 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5813 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5814 /* sv_magic() refuses to add a magic of the same 'how' as an
5817 if (how == PERL_MAGIC_taint)
5823 /* Force pos to be stored as characters, not bytes. */
5824 if (SvMAGICAL(sv) && DO_UTF8(sv)
5825 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5827 && mg->mg_flags & MGf_BYTES) {
5828 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5830 mg->mg_flags &= ~MGf_BYTES;
5833 /* Rest of work is done else where */
5834 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5837 case PERL_MAGIC_taint:
5840 case PERL_MAGIC_ext:
5841 case PERL_MAGIC_dbfile:
5848 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5855 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5857 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5858 for (mg = *mgp; mg; mg = *mgp) {
5859 const MGVTBL* const virt = mg->mg_virtual;
5860 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5861 *mgp = mg->mg_moremagic;
5862 if (virt && virt->svt_free)
5863 virt->svt_free(aTHX_ sv, mg);
5864 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5866 Safefree(mg->mg_ptr);
5867 else if (mg->mg_len == HEf_SVKEY)
5868 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5869 else if (mg->mg_type == PERL_MAGIC_utf8)
5870 Safefree(mg->mg_ptr);
5872 if (mg->mg_flags & MGf_REFCOUNTED)
5873 SvREFCNT_dec(mg->mg_obj);
5877 mgp = &mg->mg_moremagic;
5880 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5881 mg_magical(sv); /* else fix the flags now */
5890 =for apidoc sv_unmagic
5892 Removes all magic of type C<type> from an SV.
5898 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5900 PERL_ARGS_ASSERT_SV_UNMAGIC;
5901 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5905 =for apidoc sv_unmagicext
5907 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5913 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5915 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5916 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5920 =for apidoc sv_rvweaken
5922 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5923 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5924 push a back-reference to this RV onto the array of backreferences
5925 associated with that magic. If the RV is magical, set magic will be
5926 called after the RV is cleared. Silently ignores C<undef> and warns
5927 on already-weak references.
5933 Perl_sv_rvweaken(pTHX_ SV *const sv)
5937 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5939 if (!SvOK(sv)) /* let undefs pass */
5942 Perl_croak(aTHX_ "Can't weaken a nonreference");
5943 else if (SvWEAKREF(sv)) {
5944 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5947 else if (SvREADONLY(sv)) croak_no_modify();
5949 Perl_sv_add_backref(aTHX_ tsv, sv);
5951 SvREFCNT_dec_NN(tsv);
5956 =for apidoc sv_rvunweaken
5958 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5959 the backreference to this RV from the array of backreferences
5960 associated with the target SV, increment the refcount of the target.
5961 Silently ignores C<undef> and warns on non-weak references.
5967 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5971 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5973 if (!SvOK(sv)) /* let undefs pass */
5976 Perl_croak(aTHX_ "Can't unweaken a nonreference");
5977 else if (!SvWEAKREF(sv)) {
5978 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
5981 else if (SvREADONLY(sv)) croak_no_modify();
5986 SvREFCNT_inc_NN(tsv);
5987 Perl_sv_del_backref(aTHX_ tsv, sv);
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 and/or replaces a string at the specified offset/length within the SV.
6328 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6329 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6330 C<offset>. Handles get magic.
6332 =for apidoc sv_insert_flags
6334 Same as C<sv_insert>, but the extra C<flags> are passed to the
6335 C<SvPV_force_flags> that applies to C<bigstr>.
6341 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6347 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6350 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6352 SvPV_force_flags(bigstr, curlen, flags);
6353 (void)SvPOK_only_UTF8(bigstr);
6355 if (little >= SvPVX(bigstr) &&
6356 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6357 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6358 or little...little+littlelen might overlap offset...offset+len we make a copy
6360 little = savepvn(little, littlelen);
6364 if (offset + len > curlen) {
6365 SvGROW(bigstr, offset+len+1);
6366 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6367 SvCUR_set(bigstr, offset+len);
6371 i = littlelen - len;
6372 if (i > 0) { /* string might grow */
6373 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6374 mid = big + offset + len;
6375 midend = bigend = big + SvCUR(bigstr);
6378 while (midend > mid) /* shove everything down */
6379 *--bigend = *--midend;
6380 Move(little,big+offset,littlelen,char);
6381 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6386 Move(little,SvPVX(bigstr)+offset,len,char);
6391 big = SvPVX(bigstr);
6394 bigend = big + SvCUR(bigstr);
6396 if (midend > bigend)
6397 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6400 if (mid - big > bigend - midend) { /* faster to shorten from end */
6402 Move(little, mid, littlelen,char);
6405 i = bigend - midend;
6407 Move(midend, mid, i,char);
6411 SvCUR_set(bigstr, mid - big);
6413 else if ((i = mid - big)) { /* faster from front */
6414 midend -= littlelen;
6416 Move(big, midend - i, i, char);
6417 sv_chop(bigstr,midend-i);
6419 Move(little, mid, littlelen,char);
6421 else if (littlelen) {
6422 midend -= littlelen;
6423 sv_chop(bigstr,midend);
6424 Move(little,midend,littlelen,char);
6427 sv_chop(bigstr,midend);
6433 =for apidoc sv_replace
6435 Make the first argument a copy of the second, then delete the original.
6436 The target SV physically takes over ownership of the body of the source SV
6437 and inherits its flags; however, the target keeps any magic it owns,
6438 and any magic in the source is discarded.
6439 Note that this is a rather specialist SV copying operation; most of the
6440 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6446 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6448 const U32 refcnt = SvREFCNT(sv);
6450 PERL_ARGS_ASSERT_SV_REPLACE;
6452 SV_CHECK_THINKFIRST_COW_DROP(sv);
6453 if (SvREFCNT(nsv) != 1) {
6454 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6455 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6457 if (SvMAGICAL(sv)) {
6461 sv_upgrade(nsv, SVt_PVMG);
6462 SvMAGIC_set(nsv, SvMAGIC(sv));
6463 SvFLAGS(nsv) |= SvMAGICAL(sv);
6465 SvMAGIC_set(sv, NULL);
6469 assert(!SvREFCNT(sv));
6470 #ifdef DEBUG_LEAKING_SCALARS
6471 sv->sv_flags = nsv->sv_flags;
6472 sv->sv_any = nsv->sv_any;
6473 sv->sv_refcnt = nsv->sv_refcnt;
6474 sv->sv_u = nsv->sv_u;
6476 StructCopy(nsv,sv,SV);
6478 if(SvTYPE(sv) == SVt_IV) {
6479 SET_SVANY_FOR_BODYLESS_IV(sv);
6483 SvREFCNT(sv) = refcnt;
6484 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6489 /* We're about to free a GV which has a CV that refers back to us.
6490 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6494 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6499 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6502 assert(SvREFCNT(gv) == 0);
6503 assert(isGV(gv) && isGV_with_GP(gv));
6505 assert(!CvANON(cv));
6506 assert(CvGV(cv) == gv);
6507 assert(!CvNAMED(cv));
6509 /* will the CV shortly be freed by gp_free() ? */
6510 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6511 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6515 /* if not, anonymise: */
6516 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6517 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6518 : newSVpvn_flags( "__ANON__", 8, 0 );
6519 sv_catpvs(gvname, "::__ANON__");
6520 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6521 SvREFCNT_dec_NN(gvname);
6525 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6530 =for apidoc sv_clear
6532 Clear an SV: call any destructors, free up any memory used by the body,
6533 and free the body itself. The SV's head is I<not> freed, although
6534 its type is set to all 1's so that it won't inadvertently be assumed
6535 to be live during global destruction etc.
6536 This function should only be called when C<REFCNT> is zero. Most of the time
6537 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6544 Perl_sv_clear(pTHX_ SV *const orig_sv)
6549 const struct body_details *sv_type_details;
6553 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6554 Not strictly necessary */
6556 PERL_ARGS_ASSERT_SV_CLEAR;
6558 /* within this loop, sv is the SV currently being freed, and
6559 * iter_sv is the most recent AV or whatever that's being iterated
6560 * over to provide more SVs */
6566 assert(SvREFCNT(sv) == 0);
6567 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6569 if (type <= SVt_IV) {
6570 /* See the comment in sv.h about the collusion between this
6571 * early return and the overloading of the NULL slots in the
6575 SvFLAGS(sv) &= SVf_BREAK;
6576 SvFLAGS(sv) |= SVTYPEMASK;
6580 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6581 for another purpose */
6582 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6584 if (type >= SVt_PVMG) {
6586 if (!curse(sv, 1)) goto get_next_sv;
6587 type = SvTYPE(sv); /* destructor may have changed it */
6589 /* Free back-references before magic, in case the magic calls
6590 * Perl code that has weak references to sv. */
6591 if (type == SVt_PVHV) {
6592 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6596 else if (SvMAGIC(sv)) {
6597 /* Free back-references before other types of magic. */
6598 sv_unmagic(sv, PERL_MAGIC_backref);
6604 /* case SVt_INVLIST: */
6607 IoIFP(sv) != PerlIO_stdin() &&
6608 IoIFP(sv) != PerlIO_stdout() &&
6609 IoIFP(sv) != PerlIO_stderr() &&
6610 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6612 io_close(MUTABLE_IO(sv), NULL, FALSE,
6613 (IoTYPE(sv) == IoTYPE_WRONLY ||
6614 IoTYPE(sv) == IoTYPE_RDWR ||
6615 IoTYPE(sv) == IoTYPE_APPEND));
6617 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6618 PerlDir_close(IoDIRP(sv));
6619 IoDIRP(sv) = (DIR*)NULL;
6620 Safefree(IoTOP_NAME(sv));
6621 Safefree(IoFMT_NAME(sv));
6622 Safefree(IoBOTTOM_NAME(sv));
6623 if ((const GV *)sv == PL_statgv)
6627 /* FIXME for plugins */
6628 pregfree2((REGEXP*) sv);
6632 cv_undef(MUTABLE_CV(sv));
6633 /* If we're in a stash, we don't own a reference to it.
6634 * However it does have a back reference to us, which needs to
6636 if ((stash = CvSTASH(sv)))
6637 sv_del_backref(MUTABLE_SV(stash), sv);
6640 if (PL_last_swash_hv == (const HV *)sv) {
6641 PL_last_swash_hv = NULL;
6643 if (HvTOTALKEYS((HV*)sv) > 0) {
6645 /* this statement should match the one at the beginning of
6646 * hv_undef_flags() */
6647 if ( PL_phase != PERL_PHASE_DESTRUCT
6648 && (hek = HvNAME_HEK((HV*)sv)))
6650 if (PL_stashcache) {
6651 DEBUG_o(Perl_deb(aTHX_
6652 "sv_clear clearing PL_stashcache for '%" HEKf
6655 (void)hv_deletehek(PL_stashcache,
6658 hv_name_set((HV*)sv, NULL, 0, 0);
6661 /* save old iter_sv in unused SvSTASH field */
6662 assert(!SvOBJECT(sv));
6663 SvSTASH(sv) = (HV*)iter_sv;
6666 /* save old hash_index in unused SvMAGIC field */
6667 assert(!SvMAGICAL(sv));
6668 assert(!SvMAGIC(sv));
6669 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6672 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6673 goto get_next_sv; /* process this new sv */
6675 /* free empty hash */
6676 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6677 assert(!HvARRAY((HV*)sv));
6681 AV* av = MUTABLE_AV(sv);
6682 if (PL_comppad == av) {
6686 if (AvREAL(av) && AvFILLp(av) > -1) {
6687 next_sv = AvARRAY(av)[AvFILLp(av)--];
6688 /* save old iter_sv in top-most slot of AV,
6689 * and pray that it doesn't get wiped in the meantime */
6690 AvARRAY(av)[AvMAX(av)] = iter_sv;
6692 goto get_next_sv; /* process this new sv */
6694 Safefree(AvALLOC(av));
6699 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6700 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6701 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6702 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6704 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6705 SvREFCNT_dec(LvTARG(sv));
6707 /* SvLEN points to a regex body. Free the body, then
6708 * set SvLEN to whatever value was in the now-freed
6709 * regex body. The PVX buffer is shared by multiple re's
6710 * and only freed once, by the re whose len in non-null */
6711 STRLEN len = ReANY(sv)->xpv_len;
6712 pregfree2((REGEXP*) sv);
6713 SvLEN_set((sv), len);
6718 if (isGV_with_GP(sv)) {
6719 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6720 && HvENAME_get(stash))
6721 mro_method_changed_in(stash);
6722 gp_free(MUTABLE_GV(sv));
6724 unshare_hek(GvNAME_HEK(sv));
6725 /* If we're in a stash, we don't own a reference to it.
6726 * However it does have a back reference to us, which
6727 * needs to be cleared. */
6728 if ((stash = GvSTASH(sv)))
6729 sv_del_backref(MUTABLE_SV(stash), sv);
6731 /* FIXME. There are probably more unreferenced pointers to SVs
6732 * in the interpreter struct that we should check and tidy in
6733 * a similar fashion to this: */
6734 /* See also S_sv_unglob, which does the same thing. */
6735 if ((const GV *)sv == PL_last_in_gv)
6736 PL_last_in_gv = NULL;
6737 else if ((const GV *)sv == PL_statgv)
6739 else if ((const GV *)sv == PL_stderrgv)
6748 /* Don't bother with SvOOK_off(sv); as we're only going to
6752 SvOOK_offset(sv, offset);
6753 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6754 /* Don't even bother with turning off the OOK flag. */
6759 SV * const target = SvRV(sv);
6761 sv_del_backref(target, sv);
6767 else if (SvPVX_const(sv)
6768 && !(SvTYPE(sv) == SVt_PVIO
6769 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6774 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6779 if (CowREFCNT(sv)) {
6786 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6791 Safefree(SvPVX_mutable(sv));
6795 else if (SvPVX_const(sv) && SvLEN(sv)
6796 && !(SvTYPE(sv) == SVt_PVIO
6797 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6798 Safefree(SvPVX_mutable(sv));
6799 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6800 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6810 SvFLAGS(sv) &= SVf_BREAK;
6811 SvFLAGS(sv) |= SVTYPEMASK;
6813 sv_type_details = bodies_by_type + type;
6814 if (sv_type_details->arena) {
6815 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6816 &PL_body_roots[type]);
6818 else if (sv_type_details->body_size) {
6819 safefree(SvANY(sv));
6823 /* caller is responsible for freeing the head of the original sv */
6824 if (sv != orig_sv && !SvREFCNT(sv))
6827 /* grab and free next sv, if any */
6835 else if (!iter_sv) {
6837 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6838 AV *const av = (AV*)iter_sv;
6839 if (AvFILLp(av) > -1) {
6840 sv = AvARRAY(av)[AvFILLp(av)--];
6842 else { /* no more elements of current AV to free */
6845 /* restore previous value, squirrelled away */
6846 iter_sv = AvARRAY(av)[AvMAX(av)];
6847 Safefree(AvALLOC(av));
6850 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6851 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6852 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6853 /* no more elements of current HV to free */
6856 /* Restore previous values of iter_sv and hash_index,
6857 * squirrelled away */
6858 assert(!SvOBJECT(sv));
6859 iter_sv = (SV*)SvSTASH(sv);
6860 assert(!SvMAGICAL(sv));
6861 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6863 /* perl -DA does not like rubbish in SvMAGIC. */
6867 /* free any remaining detritus from the hash struct */
6868 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6869 assert(!HvARRAY((HV*)sv));
6874 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6878 if (!SvREFCNT(sv)) {
6882 if (--(SvREFCNT(sv)))
6886 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6887 "Attempt to free temp prematurely: SV 0x%" UVxf
6888 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6892 if (SvIMMORTAL(sv)) {
6893 /* make sure SvREFCNT(sv)==0 happens very seldom */
6894 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6903 /* This routine curses the sv itself, not the object referenced by sv. So
6904 sv does not have to be ROK. */
6907 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6908 PERL_ARGS_ASSERT_CURSE;
6909 assert(SvOBJECT(sv));
6911 if (PL_defstash && /* Still have a symbol table? */
6917 stash = SvSTASH(sv);
6918 assert(SvTYPE(stash) == SVt_PVHV);
6919 if (HvNAME(stash)) {
6920 CV* destructor = NULL;
6921 struct mro_meta *meta;
6923 assert (SvOOK(stash));
6925 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6928 /* don't make this an initialization above the assert, since it needs
6930 meta = HvMROMETA(stash);
6931 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6932 destructor = meta->destroy;
6933 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6934 (void *)destructor, HvNAME(stash)) );
6937 bool autoload = FALSE;
6939 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6941 destructor = GvCV(gv);
6943 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6944 GV_AUTOLOAD_ISMETHOD);
6946 destructor = GvCV(gv);
6950 /* we don't cache AUTOLOAD for DESTROY, since this code
6951 would then need to set $__PACKAGE__::AUTOLOAD, or the
6952 equivalent for XS AUTOLOADs */
6954 meta->destroy_gen = PL_sub_generation;
6955 meta->destroy = destructor;
6957 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6958 (void *)destructor, HvNAME(stash)) );
6961 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6965 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6967 /* A constant subroutine can have no side effects, so
6968 don't bother calling it. */
6969 && !CvCONST(destructor)
6970 /* Don't bother calling an empty destructor or one that
6971 returns immediately. */
6972 && (CvISXSUB(destructor)
6973 || (CvSTART(destructor)
6974 && (CvSTART(destructor)->op_next->op_type
6976 && (CvSTART(destructor)->op_next->op_type
6978 || CvSTART(destructor)->op_next->op_next->op_type
6984 SV* const tmpref = newRV(sv);
6985 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6987 PUSHSTACKi(PERLSI_DESTROY);
6992 call_sv(MUTABLE_SV(destructor),
6993 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6997 if(SvREFCNT(tmpref) < 2) {
6998 /* tmpref is not kept alive! */
7000 SvRV_set(tmpref, NULL);
7003 SvREFCNT_dec_NN(tmpref);
7006 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7009 if (check_refcnt && SvREFCNT(sv)) {
7010 if (PL_in_clean_objs)
7012 "DESTROY created new reference to dead object '%" HEKf "'",
7013 HEKfARG(HvNAME_HEK(stash)));
7014 /* DESTROY gave object new lease on life */
7020 HV * const stash = SvSTASH(sv);
7021 /* Curse before freeing the stash, as freeing the stash could cause
7022 a recursive call into S_curse. */
7023 SvOBJECT_off(sv); /* Curse the object. */
7024 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7025 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7031 =for apidoc sv_newref
7033 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7040 Perl_sv_newref(pTHX_ SV *const sv)
7042 PERL_UNUSED_CONTEXT;
7051 Decrement an SV's reference count, and if it drops to zero, call
7052 C<sv_clear> to invoke destructors and free up any memory used by
7053 the body; finally, deallocating the SV's head itself.
7054 Normally called via a wrapper macro C<SvREFCNT_dec>.
7060 Perl_sv_free(pTHX_ SV *const sv)
7066 /* Private helper function for SvREFCNT_dec().
7067 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7070 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7074 PERL_ARGS_ASSERT_SV_FREE2;
7076 if (LIKELY( rc == 1 )) {
7082 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7083 "Attempt to free temp prematurely: SV 0x%" UVxf
7084 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7088 if (SvIMMORTAL(sv)) {
7089 /* make sure SvREFCNT(sv)==0 happens very seldom */
7090 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7094 if (! SvREFCNT(sv)) /* may have have been resurrected */
7099 /* handle exceptional cases */
7103 if (SvFLAGS(sv) & SVf_BREAK)
7104 /* this SV's refcnt has been artificially decremented to
7105 * trigger cleanup */
7107 if (PL_in_clean_all) /* All is fair */
7109 if (SvIMMORTAL(sv)) {
7110 /* make sure SvREFCNT(sv)==0 happens very seldom */
7111 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7114 if (ckWARN_d(WARN_INTERNAL)) {
7115 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7116 Perl_dump_sv_child(aTHX_ sv);
7118 #ifdef DEBUG_LEAKING_SCALARS
7121 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7122 if (PL_warnhook == PERL_WARNHOOK_FATAL
7123 || ckDEAD(packWARN(WARN_INTERNAL))) {
7124 /* Don't let Perl_warner cause us to escape our fate: */
7128 /* This may not return: */
7129 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7130 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7131 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7134 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7144 Returns the length of the string in the SV. Handles magic and type
7145 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7146 gives raw access to the C<xpv_cur> slot.
7152 Perl_sv_len(pTHX_ SV *const sv)
7159 (void)SvPV_const(sv, len);
7164 =for apidoc sv_len_utf8
7166 Returns the number of characters in the string in an SV, counting wide
7167 UTF-8 bytes as a single character. Handles magic and type coercion.
7173 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7174 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7175 * (Note that the mg_len is not the length of the mg_ptr field.
7176 * This allows the cache to store the character length of the string without
7177 * needing to malloc() extra storage to attach to the mg_ptr.)
7182 Perl_sv_len_utf8(pTHX_ SV *const sv)
7188 return sv_len_utf8_nomg(sv);
7192 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7195 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7197 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7199 if (PL_utf8cache && SvUTF8(sv)) {
7201 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7203 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7204 if (mg->mg_len != -1)
7207 /* We can use the offset cache for a headstart.
7208 The longer value is stored in the first pair. */
7209 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7211 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7215 if (PL_utf8cache < 0) {
7216 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7217 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7221 ulen = Perl_utf8_length(aTHX_ s, s + len);
7222 utf8_mg_len_cache_update(sv, &mg, ulen);
7226 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7229 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7232 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7233 STRLEN *const uoffset_p, bool *const at_end)
7235 const U8 *s = start;
7236 STRLEN uoffset = *uoffset_p;
7238 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7240 while (s < send && uoffset) {
7247 else if (s > send) {
7249 /* This is the existing behaviour. Possibly it should be a croak, as
7250 it's actually a bounds error */
7253 *uoffset_p -= uoffset;
7257 /* Given the length of the string in both bytes and UTF-8 characters, decide
7258 whether to walk forwards or backwards to find the byte corresponding to
7259 the passed in UTF-8 offset. */
7261 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7262 STRLEN uoffset, const STRLEN uend)
7264 STRLEN backw = uend - uoffset;
7266 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7268 if (uoffset < 2 * backw) {
7269 /* The assumption is that going forwards is twice the speed of going
7270 forward (that's where the 2 * backw comes from).
7271 (The real figure of course depends on the UTF-8 data.) */
7272 const U8 *s = start;
7274 while (s < send && uoffset--)
7284 while (UTF8_IS_CONTINUATION(*send))
7287 return send - start;
7290 /* For the string representation of the given scalar, find the byte
7291 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7292 give another position in the string, *before* the sought offset, which
7293 (which is always true, as 0, 0 is a valid pair of positions), which should
7294 help reduce the amount of linear searching.
7295 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7296 will be used to reduce the amount of linear searching. The cache will be
7297 created if necessary, and the found value offered to it for update. */
7299 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7300 const U8 *const send, STRLEN uoffset,
7301 STRLEN uoffset0, STRLEN boffset0)
7303 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7305 bool at_end = FALSE;
7307 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7309 assert (uoffset >= uoffset0);
7314 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7316 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7317 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7318 if ((*mgp)->mg_ptr) {
7319 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7320 if (cache[0] == uoffset) {
7321 /* An exact match. */
7324 if (cache[2] == uoffset) {
7325 /* An exact match. */
7329 if (cache[0] < uoffset) {
7330 /* The cache already knows part of the way. */
7331 if (cache[0] > uoffset0) {
7332 /* The cache knows more than the passed in pair */
7333 uoffset0 = cache[0];
7334 boffset0 = cache[1];
7336 if ((*mgp)->mg_len != -1) {
7337 /* And we know the end too. */
7339 + sv_pos_u2b_midway(start + boffset0, send,
7341 (*mgp)->mg_len - uoffset0);
7343 uoffset -= uoffset0;
7345 + sv_pos_u2b_forwards(start + boffset0,
7346 send, &uoffset, &at_end);
7347 uoffset += uoffset0;
7350 else if (cache[2] < uoffset) {
7351 /* We're between the two cache entries. */
7352 if (cache[2] > uoffset0) {
7353 /* and the cache knows more than the passed in pair */
7354 uoffset0 = cache[2];
7355 boffset0 = cache[3];
7359 + sv_pos_u2b_midway(start + boffset0,
7362 cache[0] - uoffset0);
7365 + sv_pos_u2b_midway(start + boffset0,
7368 cache[2] - uoffset0);
7372 else if ((*mgp)->mg_len != -1) {
7373 /* If we can take advantage of a passed in offset, do so. */
7374 /* In fact, offset0 is either 0, or less than offset, so don't
7375 need to worry about the other possibility. */
7377 + sv_pos_u2b_midway(start + boffset0, send,
7379 (*mgp)->mg_len - uoffset0);
7384 if (!found || PL_utf8cache < 0) {
7385 STRLEN real_boffset;
7386 uoffset -= uoffset0;
7387 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7388 send, &uoffset, &at_end);
7389 uoffset += uoffset0;
7391 if (found && PL_utf8cache < 0)
7392 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7394 boffset = real_boffset;
7397 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7399 utf8_mg_len_cache_update(sv, mgp, uoffset);
7401 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7408 =for apidoc sv_pos_u2b_flags
7410 Converts the offset from a count of UTF-8 chars from
7411 the start of the string, to a count of the equivalent number of bytes; if
7412 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7413 C<offset>, rather than from the start
7414 of the string. Handles type coercion.
7415 C<flags> is passed to C<SvPV_flags>, and usually should be
7416 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7422 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7423 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7424 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7429 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7436 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7438 start = (U8*)SvPV_flags(sv, len, flags);
7440 const U8 * const send = start + len;
7442 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7445 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7446 is 0, and *lenp is already set to that. */) {
7447 /* Convert the relative offset to absolute. */
7448 const STRLEN uoffset2 = uoffset + *lenp;
7449 const STRLEN boffset2
7450 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7451 uoffset, boffset) - boffset;
7465 =for apidoc sv_pos_u2b
7467 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7468 the start of the string, to a count of the equivalent number of bytes; if
7469 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7470 the offset, rather than from the start of the string. Handles magic and
7473 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7480 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7481 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7482 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7486 /* This function is subject to size and sign problems */
7489 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7491 PERL_ARGS_ASSERT_SV_POS_U2B;
7494 STRLEN ulen = (STRLEN)*lenp;
7495 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7496 SV_GMAGIC|SV_CONST_RETURN);
7499 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7500 SV_GMAGIC|SV_CONST_RETURN);
7505 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7508 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7509 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7512 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7513 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7514 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7518 (*mgp)->mg_len = ulen;
7521 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7522 byte length pairing. The (byte) length of the total SV is passed in too,
7523 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7524 may not have updated SvCUR, so we can't rely on reading it directly.
7526 The proffered utf8/byte length pairing isn't used if the cache already has
7527 two pairs, and swapping either for the proffered pair would increase the
7528 RMS of the intervals between known byte offsets.
7530 The cache itself consists of 4 STRLEN values
7531 0: larger UTF-8 offset
7532 1: corresponding byte offset
7533 2: smaller UTF-8 offset
7534 3: corresponding byte offset
7536 Unused cache pairs have the value 0, 0.
7537 Keeping the cache "backwards" means that the invariant of
7538 cache[0] >= cache[2] is maintained even with empty slots, which means that
7539 the code that uses it doesn't need to worry if only 1 entry has actually
7540 been set to non-zero. It also makes the "position beyond the end of the
7541 cache" logic much simpler, as the first slot is always the one to start
7545 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7546 const STRLEN utf8, const STRLEN blen)
7550 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7555 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7556 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7557 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7559 (*mgp)->mg_len = -1;
7563 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7564 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7565 (*mgp)->mg_ptr = (char *) cache;
7569 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7570 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7571 a pointer. Note that we no longer cache utf8 offsets on refer-
7572 ences, but this check is still a good idea, for robustness. */
7573 const U8 *start = (const U8 *) SvPVX_const(sv);
7574 const STRLEN realutf8 = utf8_length(start, start + byte);
7576 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7580 /* Cache is held with the later position first, to simplify the code
7581 that deals with unbounded ends. */
7583 ASSERT_UTF8_CACHE(cache);
7584 if (cache[1] == 0) {
7585 /* Cache is totally empty */
7588 } else if (cache[3] == 0) {
7589 if (byte > cache[1]) {
7590 /* New one is larger, so goes first. */
7591 cache[2] = cache[0];
7592 cache[3] = cache[1];
7600 /* float casts necessary? XXX */
7601 #define THREEWAY_SQUARE(a,b,c,d) \
7602 ((float)((d) - (c))) * ((float)((d) - (c))) \
7603 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7604 + ((float)((b) - (a))) * ((float)((b) - (a)))
7606 /* Cache has 2 slots in use, and we know three potential pairs.
7607 Keep the two that give the lowest RMS distance. Do the
7608 calculation in bytes simply because we always know the byte
7609 length. squareroot has the same ordering as the positive value,
7610 so don't bother with the actual square root. */
7611 if (byte > cache[1]) {
7612 /* New position is after the existing pair of pairs. */
7613 const float keep_earlier
7614 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7615 const float keep_later
7616 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7618 if (keep_later < keep_earlier) {
7619 cache[2] = cache[0];
7620 cache[3] = cache[1];
7626 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7627 float b, c, keep_earlier;
7628 if (byte > cache[3]) {
7629 /* New position is between the existing pair of pairs. */
7630 b = (float)cache[3];
7633 /* New position is before the existing pair of pairs. */
7635 c = (float)cache[3];
7637 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7638 if (byte > cache[3]) {
7639 if (keep_later < keep_earlier) {
7649 if (! (keep_later < keep_earlier)) {
7650 cache[0] = cache[2];
7651 cache[1] = cache[3];
7658 ASSERT_UTF8_CACHE(cache);
7661 /* We already know all of the way, now we may be able to walk back. The same
7662 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7663 backward is half the speed of walking forward. */
7665 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7666 const U8 *end, STRLEN endu)
7668 const STRLEN forw = target - s;
7669 STRLEN backw = end - target;
7671 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7673 if (forw < 2 * backw) {
7674 return utf8_length(s, target);
7677 while (end > target) {
7679 while (UTF8_IS_CONTINUATION(*end)) {
7688 =for apidoc sv_pos_b2u_flags
7690 Converts C<offset> from a count of bytes from the start of the string, to
7691 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7692 C<flags> is passed to C<SvPV_flags>, and usually should be
7693 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7699 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7700 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7705 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7708 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7714 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7716 s = (const U8*)SvPV_flags(sv, blen, flags);
7719 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7720 ", byte=%" UVuf, (UV)blen, (UV)offset);
7726 && SvTYPE(sv) >= SVt_PVMG
7727 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7730 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7731 if (cache[1] == offset) {
7732 /* An exact match. */
7735 if (cache[3] == offset) {
7736 /* An exact match. */
7740 if (cache[1] < offset) {
7741 /* We already know part of the way. */
7742 if (mg->mg_len != -1) {
7743 /* Actually, we know the end too. */
7745 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7746 s + blen, mg->mg_len - cache[0]);
7748 len = cache[0] + utf8_length(s + cache[1], send);
7751 else if (cache[3] < offset) {
7752 /* We're between the two cached pairs, so we do the calculation
7753 offset by the byte/utf-8 positions for the earlier pair,
7754 then add the utf-8 characters from the string start to
7756 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7757 s + cache[1], cache[0] - cache[2])
7761 else { /* cache[3] > offset */
7762 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7766 ASSERT_UTF8_CACHE(cache);
7768 } else if (mg->mg_len != -1) {
7769 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7773 if (!found || PL_utf8cache < 0) {
7774 const STRLEN real_len = utf8_length(s, send);
7776 if (found && PL_utf8cache < 0)
7777 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7783 utf8_mg_len_cache_update(sv, &mg, len);
7785 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7792 =for apidoc sv_pos_b2u
7794 Converts the value pointed to by C<offsetp> from a count of bytes from the
7795 start of the string, to a count of the equivalent number of UTF-8 chars.
7796 Handles magic and type coercion.
7798 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7805 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7806 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7811 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7813 PERL_ARGS_ASSERT_SV_POS_B2U;
7818 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7819 SV_GMAGIC|SV_CONST_RETURN);
7823 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7824 STRLEN real, SV *const sv)
7826 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7828 /* As this is debugging only code, save space by keeping this test here,
7829 rather than inlining it in all the callers. */
7830 if (from_cache == real)
7833 /* Need to turn the assertions off otherwise we may recurse infinitely
7834 while printing error messages. */
7835 SAVEI8(PL_utf8cache);
7837 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7838 func, (UV) from_cache, (UV) real, SVfARG(sv));
7844 Returns a boolean indicating whether the strings in the two SVs are
7845 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7846 coerce its args to strings if necessary.
7848 =for apidoc sv_eq_flags
7850 Returns a boolean indicating whether the strings in the two SVs are
7851 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7852 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7858 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7870 /* if pv1 and pv2 are the same, second SvPV_const call may
7871 * invalidate pv1 (if we are handling magic), so we may need to
7873 if (sv1 == sv2 && flags & SV_GMAGIC
7874 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7875 pv1 = SvPV_const(sv1, cur1);
7876 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7878 pv1 = SvPV_flags_const(sv1, cur1, flags);
7886 pv2 = SvPV_flags_const(sv2, cur2, flags);
7888 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7889 /* Differing utf8ness. */
7891 /* sv1 is the UTF-8 one */
7892 return bytes_cmp_utf8((const U8*)pv2, cur2,
7893 (const U8*)pv1, cur1) == 0;
7896 /* sv2 is the UTF-8 one */
7897 return bytes_cmp_utf8((const U8*)pv1, cur1,
7898 (const U8*)pv2, cur2) == 0;
7903 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7911 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7912 string in C<sv1> is less than, equal to, or greater than the string in
7913 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7914 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7916 =for apidoc sv_cmp_flags
7918 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7919 string in C<sv1> is less than, equal to, or greater than the string in
7920 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7921 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7922 also C<L</sv_cmp_locale_flags>>.
7928 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7930 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7934 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7938 const char *pv1, *pv2;
7940 SV *svrecode = NULL;
7947 pv1 = SvPV_flags_const(sv1, cur1, flags);
7954 pv2 = SvPV_flags_const(sv2, cur2, flags);
7956 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7957 /* Differing utf8ness. */
7959 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7960 (const U8*)pv1, cur1);
7961 return retval ? retval < 0 ? -1 : +1 : 0;
7964 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7965 (const U8*)pv2, cur2);
7966 return retval ? retval < 0 ? -1 : +1 : 0;
7970 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7973 cmp = cur2 ? -1 : 0;
7977 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7980 if (! DO_UTF8(sv1)) {
7982 const I32 retval = memcmp((const void*)pv1,
7986 cmp = retval < 0 ? -1 : 1;
7987 } else if (cur1 == cur2) {
7990 cmp = cur1 < cur2 ? -1 : 1;
7994 else { /* Both are to be treated as UTF-EBCDIC */
7996 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7997 * which remaps code points 0-255. We therefore generally have to
7998 * unmap back to the original values to get an accurate comparison.
7999 * But we don't have to do that for UTF-8 invariants, as by
8000 * definition, they aren't remapped, nor do we have to do it for
8001 * above-latin1 code points, as they also aren't remapped. (This
8002 * code also works on ASCII platforms, but the memcmp() above is
8005 const char *e = pv1 + shortest_len;
8007 /* Find the first bytes that differ between the two strings */
8008 while (pv1 < e && *pv1 == *pv2) {
8014 if (pv1 == e) { /* Are the same all the way to the end */
8018 cmp = cur1 < cur2 ? -1 : 1;
8021 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8022 * in the strings were. The current bytes may or may not be
8023 * at the beginning of a character. But neither or both are
8024 * (or else earlier bytes would have been different). And
8025 * if we are in the middle of a character, the two
8026 * characters are comprised of the same number of bytes
8027 * (because in this case the start bytes are the same, and
8028 * the start bytes encode the character's length). */
8029 if (UTF8_IS_INVARIANT(*pv1))
8031 /* If both are invariants; can just compare directly */
8032 if (UTF8_IS_INVARIANT(*pv2)) {
8033 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8035 else /* Since *pv1 is invariant, it is the whole character,
8036 which means it is at the beginning of a character.
8037 That means pv2 is also at the beginning of a
8038 character (see earlier comment). Since it isn't
8039 invariant, it must be a start byte. If it starts a
8040 character whose code point is above 255, that
8041 character is greater than any single-byte char, which
8043 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8048 /* Here, pv2 points to a character composed of 2 bytes
8049 * whose code point is < 256. Get its code point and
8050 * compare with *pv1 */
8051 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8056 else /* The code point starting at pv1 isn't a single byte */
8057 if (UTF8_IS_INVARIANT(*pv2))
8059 /* But here, the code point starting at *pv2 is a single byte,
8060 * and so *pv1 must begin a character, hence is a start byte.
8061 * If that character is above 255, it is larger than any
8062 * single-byte char, which *pv2 is */
8063 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8067 /* Here, pv1 points to a character composed of 2 bytes
8068 * whose code point is < 256. Get its code point and
8069 * compare with the single byte character *pv2 */
8070 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8075 else /* Here, we've ruled out either *pv1 and *pv2 being
8076 invariant. That means both are part of variants, but not
8077 necessarily at the start of a character */
8078 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8079 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8081 /* Here, at least one is the start of a character, which means
8082 * the other is also a start byte. And the code point of at
8083 * least one of the characters is above 255. It is a
8084 * characteristic of UTF-EBCDIC that all start bytes for
8085 * above-latin1 code points are well behaved as far as code
8086 * point comparisons go, and all are larger than all other
8087 * start bytes, so the comparison with those is also well
8089 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8092 /* Here both *pv1 and *pv2 are part of variant characters.
8093 * They could be both continuations, or both start characters.
8094 * (One or both could even be an illegal start character (for
8095 * an overlong) which for the purposes of sorting we treat as
8097 if (UTF8_IS_CONTINUATION(*pv1)) {
8099 /* If they are continuations for code points above 255,
8100 * then comparing the current byte is sufficient, as there
8101 * is no remapping of these and so the comparison is
8102 * well-behaved. We determine if they are such
8103 * continuations by looking at the preceding byte. It
8104 * could be a start byte, from which we can tell if it is
8105 * for an above 255 code point. Or it could be a
8106 * continuation, which means the character occupies at
8107 * least 3 bytes, so must be above 255. */
8108 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8109 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8111 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8115 /* Here, the continuations are for code points below 256;
8116 * back up one to get to the start byte */
8121 /* We need to get the actual native code point of each of these
8122 * variants in order to compare them */
8123 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8124 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8133 SvREFCNT_dec(svrecode);
8139 =for apidoc sv_cmp_locale
8141 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8142 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8143 if necessary. See also C<L</sv_cmp>>.
8145 =for apidoc sv_cmp_locale_flags
8147 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8148 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8149 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8150 C<L</sv_cmp_flags>>.
8156 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8158 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8162 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8165 #ifdef USE_LOCALE_COLLATE
8171 if (PL_collation_standard)
8176 /* Revert to using raw compare if both operands exist, but either one
8177 * doesn't transform properly for collation */
8179 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8183 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8189 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8190 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8193 if (!pv1 || !len1) {
8204 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8207 return retval < 0 ? -1 : 1;
8210 * When the result of collation is equality, that doesn't mean
8211 * that there are no differences -- some locales exclude some
8212 * characters from consideration. So to avoid false equalities,
8213 * we use the raw string as a tiebreaker.
8220 PERL_UNUSED_ARG(flags);
8221 #endif /* USE_LOCALE_COLLATE */
8223 return sv_cmp(sv1, sv2);
8227 #ifdef USE_LOCALE_COLLATE
8230 =for apidoc sv_collxfrm
8232 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8233 C<L</sv_collxfrm_flags>>.
8235 =for apidoc sv_collxfrm_flags
8237 Add Collate Transform magic to an SV if it doesn't already have it. If the
8238 flags contain C<SV_GMAGIC>, it handles get-magic.
8240 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8241 scalar data of the variable, but transformed to such a format that a normal
8242 memory comparison can be used to compare the data according to the locale
8249 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8253 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8255 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8257 /* If we don't have collation magic on 'sv', or the locale has changed
8258 * since the last time we calculated it, get it and save it now */
8259 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8264 /* Free the old space */
8266 Safefree(mg->mg_ptr);
8268 s = SvPV_flags_const(sv, len, flags);
8269 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8271 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8286 if (mg && mg->mg_ptr) {
8288 return mg->mg_ptr + sizeof(PL_collation_ix);
8296 #endif /* USE_LOCALE_COLLATE */
8299 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8301 SV * const tsv = newSV(0);
8304 sv_gets(tsv, fp, 0);
8305 sv_utf8_upgrade_nomg(tsv);
8306 SvCUR_set(sv,append);
8309 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8313 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8316 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8317 /* Grab the size of the record we're getting */
8318 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8325 /* With a true, record-oriented file on VMS, we need to use read directly
8326 * to ensure that we respect RMS record boundaries. The user is responsible
8327 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8328 * record size) field. N.B. This is likely to produce invalid results on
8329 * varying-width character data when a record ends mid-character.
8331 fd = PerlIO_fileno(fp);
8333 && PerlLIO_fstat(fd, &st) == 0
8334 && (st.st_fab_rfm == FAB$C_VAR
8335 || st.st_fab_rfm == FAB$C_VFC
8336 || st.st_fab_rfm == FAB$C_FIX)) {
8338 bytesread = PerlLIO_read(fd, buffer, recsize);
8340 else /* in-memory file from PerlIO::Scalar
8341 * or not a record-oriented file
8345 bytesread = PerlIO_read(fp, buffer, recsize);
8347 /* At this point, the logic in sv_get() means that sv will
8348 be treated as utf-8 if the handle is utf8.
8350 if (PerlIO_isutf8(fp) && bytesread > 0) {
8351 char *bend = buffer + bytesread;
8352 char *bufp = buffer;
8353 size_t charcount = 0;
8354 bool charstart = TRUE;
8357 while (charcount < recsize) {
8358 /* count accumulated characters */
8359 while (bufp < bend) {
8361 skip = UTF8SKIP(bufp);
8363 if (bufp + skip > bend) {
8364 /* partial at the end */
8375 if (charcount < recsize) {
8377 STRLEN bufp_offset = bufp - buffer;
8378 SSize_t morebytesread;
8380 /* originally I read enough to fill any incomplete
8381 character and the first byte of the next
8382 character if needed, but if there's many
8383 multi-byte encoded characters we're going to be
8384 making a read call for every character beyond
8385 the original read size.
8387 So instead, read the rest of the character if
8388 any, and enough bytes to match at least the
8389 start bytes for each character we're going to
8393 readsize = recsize - charcount;
8395 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8396 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8397 bend = buffer + bytesread;
8398 morebytesread = PerlIO_read(fp, bend, readsize);
8399 if (morebytesread <= 0) {
8400 /* we're done, if we still have incomplete
8401 characters the check code in sv_gets() will
8404 I'd originally considered doing
8405 PerlIO_ungetc() on all but the lead
8406 character of the incomplete character, but
8407 read() doesn't do that, so I don't.
8412 /* prepare to scan some more */
8413 bytesread += morebytesread;
8414 bend = buffer + bytesread;
8415 bufp = buffer + bufp_offset;
8423 SvCUR_set(sv, bytesread + append);
8424 buffer[bytesread] = '\0';
8425 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8431 Get a line from the filehandle and store it into the SV, optionally
8432 appending to the currently-stored string. If C<append> is not 0, the
8433 line is appended to the SV instead of overwriting it. C<append> should
8434 be set to the byte offset that the appended string should start at
8435 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8441 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8451 PERL_ARGS_ASSERT_SV_GETS;
8453 if (SvTHINKFIRST(sv))
8454 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8455 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8457 However, perlbench says it's slower, because the existing swipe code
8458 is faster than copy on write.
8459 Swings and roundabouts. */
8460 SvUPGRADE(sv, SVt_PV);
8463 /* line is going to be appended to the existing buffer in the sv */
8464 if (PerlIO_isutf8(fp)) {
8466 sv_utf8_upgrade_nomg(sv);
8467 sv_pos_u2b(sv,&append,0);
8469 } else if (SvUTF8(sv)) {
8470 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8476 /* not appending - "clear" the string by setting SvCUR to 0,
8477 * the pv is still avaiable. */
8480 if (PerlIO_isutf8(fp))
8483 if (IN_PERL_COMPILETIME) {
8484 /* we always read code in line mode */
8488 else if (RsSNARF(PL_rs)) {
8489 /* If it is a regular disk file use size from stat() as estimate
8490 of amount we are going to read -- may result in mallocing
8491 more memory than we really need if the layers below reduce
8492 the size we read (e.g. CRLF or a gzip layer).
8495 int fd = PerlIO_fileno(fp);
8496 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8497 const Off_t offset = PerlIO_tell(fp);
8498 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8499 #ifdef PERL_COPY_ON_WRITE
8500 /* Add an extra byte for the sake of copy-on-write's
8501 * buffer reference count. */
8502 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8504 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8511 else if (RsRECORD(PL_rs)) {
8512 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8514 else if (RsPARA(PL_rs)) {
8520 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8521 if (PerlIO_isutf8(fp)) {
8522 rsptr = SvPVutf8(PL_rs, rslen);
8525 if (SvUTF8(PL_rs)) {
8526 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8527 Perl_croak(aTHX_ "Wide character in $/");
8530 /* extract the raw pointer to the record separator */
8531 rsptr = SvPV_const(PL_rs, rslen);
8535 /* rslast is the last character in the record separator
8536 * note we don't use rslast except when rslen is true, so the
8537 * null assign is a placeholder. */
8538 rslast = rslen ? rsptr[rslen - 1] : '\0';
8540 if (rspara) { /* have to do this both before and after */
8541 /* to make sure file boundaries work right */
8545 i = PerlIO_getc(fp);
8549 PerlIO_ungetc(fp,i);
8555 /* See if we know enough about I/O mechanism to cheat it ! */
8557 /* This used to be #ifdef test - it is made run-time test for ease
8558 of abstracting out stdio interface. One call should be cheap
8559 enough here - and may even be a macro allowing compile
8563 if (PerlIO_fast_gets(fp)) {
8565 * We can do buffer based IO operations on this filehandle.
8567 * This means we can bypass a lot of subcalls and process
8568 * the buffer directly, it also means we know the upper bound
8569 * on the amount of data we might read of the current buffer
8570 * into our sv. Knowing this allows us to preallocate the pv
8571 * to be able to hold that maximum, which allows us to simplify
8572 * a lot of logic. */
8575 * We're going to steal some values from the stdio struct
8576 * and put EVERYTHING in the innermost loop into registers.
8578 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8579 STRLEN bpx; /* length of the data in the target sv
8580 used to fix pointers after a SvGROW */
8581 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8582 of data left in the read-ahead buffer.
8583 If 0 then the pv buffer can hold the full
8584 amount left, otherwise this is the amount it
8587 /* Here is some breathtakingly efficient cheating */
8589 /* When you read the following logic resist the urge to think
8590 * of record separators that are 1 byte long. They are an
8591 * uninteresting special (simple) case.
8593 * Instead think of record separators which are at least 2 bytes
8594 * long, and keep in mind that we need to deal with such
8595 * separators when they cross a read-ahead buffer boundary.
8597 * Also consider that we need to gracefully deal with separators
8598 * that may be longer than a single read ahead buffer.
8600 * Lastly do not forget we want to copy the delimiter as well. We
8601 * are copying all data in the file _up_to_and_including_ the separator
8604 * Now that you have all that in mind here is what is happening below:
8606 * 1. When we first enter the loop we do some memory book keeping to see
8607 * how much free space there is in the target SV. (This sub assumes that
8608 * it is operating on the same SV most of the time via $_ and that it is
8609 * going to be able to reuse the same pv buffer each call.) If there is
8610 * "enough" room then we set "shortbuffered" to how much space there is
8611 * and start reading forward.
8613 * 2. When we scan forward we copy from the read-ahead buffer to the target
8614 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8615 * and the end of the of pv, as well as for the "rslast", which is the last
8616 * char of the separator.
8618 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8619 * (which has a "complete" record up to the point we saw rslast) and check
8620 * it to see if it matches the separator. If it does we are done. If it doesn't
8621 * we continue on with the scan/copy.
8623 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8624 * the IO system to read the next buffer. We do this by doing a getc(), which
8625 * returns a single char read (or EOF), and prefills the buffer, and also
8626 * allows us to find out how full the buffer is. We use this information to
8627 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8628 * the returned single char into the target sv, and then go back into scan
8631 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8632 * remaining space in the read-buffer.
8634 * Note that this code despite its twisty-turny nature is pretty darn slick.
8635 * It manages single byte separators, multi-byte cross boundary separators,
8636 * and cross-read-buffer separators cleanly and efficiently at the cost
8637 * of potentially greatly overallocating the target SV.
8643 /* get the number of bytes remaining in the read-ahead buffer
8644 * on first call on a given fp this will return 0.*/
8645 cnt = PerlIO_get_cnt(fp);
8647 /* make sure we have the room */
8648 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8649 /* Not room for all of it
8650 if we are looking for a separator and room for some
8652 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8653 /* just process what we have room for */
8654 shortbuffered = cnt - SvLEN(sv) + append + 1;
8655 cnt -= shortbuffered;
8658 /* ensure that the target sv has enough room to hold
8659 * the rest of the read-ahead buffer */
8661 /* remember that cnt can be negative */
8662 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8666 /* we have enough room to hold the full buffer, lets scream */
8670 /* extract the pointer to sv's string buffer, offset by append as necessary */
8671 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8672 /* extract the point to the read-ahead buffer */
8673 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8675 /* some trace debug output */
8676 DEBUG_P(PerlIO_printf(Perl_debug_log,
8677 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8678 DEBUG_P(PerlIO_printf(Perl_debug_log,
8679 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8681 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8682 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8686 /* if there is stuff left in the read-ahead buffer */
8688 /* if there is a separator */
8690 /* find next rslast */
8693 /* shortcut common case of blank line */
8695 if ((*bp++ = *ptr++) == rslast)
8696 goto thats_all_folks;
8698 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8700 SSize_t got = p - ptr + 1;
8701 Copy(ptr, bp, got, STDCHAR);
8705 goto thats_all_folks;
8707 Copy(ptr, bp, cnt, STDCHAR);
8713 /* no separator, slurp the full buffer */
8714 Copy(ptr, bp, cnt, char); /* this | eat */
8715 bp += cnt; /* screams | dust */
8716 ptr += cnt; /* louder | sed :-) */
8718 assert (!shortbuffered);
8719 goto cannot_be_shortbuffered;
8723 if (shortbuffered) { /* oh well, must extend */
8724 /* we didnt have enough room to fit the line into the target buffer
8725 * so we must extend the target buffer and keep going */
8726 cnt = shortbuffered;
8728 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8730 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8731 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8732 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8736 cannot_be_shortbuffered:
8737 /* we need to refill the read-ahead buffer if possible */
8739 DEBUG_P(PerlIO_printf(Perl_debug_log,
8740 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8741 PTR2UV(ptr),(IV)cnt));
8742 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8744 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8745 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8746 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8747 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8750 call PerlIO_getc() to let it prefill the lookahead buffer
8752 This used to call 'filbuf' in stdio form, but as that behaves like
8753 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8754 another abstraction.
8756 Note we have to deal with the char in 'i' if we are not at EOF
8758 bpx = bp - (STDCHAR*)SvPVX_const(sv);
8759 /* signals might be called here, possibly modifying sv */
8760 i = PerlIO_getc(fp); /* get more characters */
8761 bp = (STDCHAR*)SvPVX_const(sv) + bpx;
8763 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8764 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8765 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8766 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8768 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8769 cnt = PerlIO_get_cnt(fp);
8770 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8771 DEBUG_P(PerlIO_printf(Perl_debug_log,
8772 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8773 PTR2UV(ptr),(IV)cnt));
8775 if (i == EOF) /* all done for ever? */
8776 goto thats_really_all_folks;
8778 /* make sure we have enough space in the target sv */
8779 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8781 SvGROW(sv, bpx + cnt + 2);
8782 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8784 /* copy of the char we got from getc() */
8785 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8787 /* make sure we deal with the i being the last character of a separator */
8788 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8789 goto thats_all_folks;
8793 /* check if we have actually found the separator - only really applies
8795 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8796 memNE((char*)bp - rslen, rsptr, rslen))
8797 goto screamer; /* go back to the fray */
8798 thats_really_all_folks:
8800 cnt += shortbuffered;
8801 DEBUG_P(PerlIO_printf(Perl_debug_log,
8802 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8803 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8804 DEBUG_P(PerlIO_printf(Perl_debug_log,
8805 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8807 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8808 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8810 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8811 DEBUG_P(PerlIO_printf(Perl_debug_log,
8812 "Screamer: done, len=%ld, string=|%.*s|\n",
8813 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8817 /*The big, slow, and stupid way. */
8818 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8819 STDCHAR *buf = NULL;
8820 Newx(buf, 8192, STDCHAR);
8828 const STDCHAR * const bpe = buf + sizeof(buf);
8830 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8831 ; /* keep reading */
8835 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8836 /* Accommodate broken VAXC compiler, which applies U8 cast to
8837 * both args of ?: operator, causing EOF to change into 255
8840 i = (U8)buf[cnt - 1];
8846 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8848 sv_catpvn_nomg(sv, (char *) buf, cnt);
8850 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8852 if (i != EOF && /* joy */
8854 SvCUR(sv) < rslen ||
8855 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8859 * If we're reading from a TTY and we get a short read,
8860 * indicating that the user hit his EOF character, we need
8861 * to notice it now, because if we try to read from the TTY
8862 * again, the EOF condition will disappear.
8864 * The comparison of cnt to sizeof(buf) is an optimization
8865 * that prevents unnecessary calls to feof().
8869 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8873 #ifdef USE_HEAP_INSTEAD_OF_STACK
8878 if (rspara) { /* have to do this both before and after */
8879 while (i != EOF) { /* to make sure file boundaries work right */
8880 i = PerlIO_getc(fp);
8882 PerlIO_ungetc(fp,i);
8888 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8894 Auto-increment of the value in the SV, doing string to numeric conversion
8895 if necessary. Handles 'get' magic and operator overloading.
8901 Perl_sv_inc(pTHX_ SV *const sv)
8910 =for apidoc sv_inc_nomg
8912 Auto-increment of the value in the SV, doing string to numeric conversion
8913 if necessary. Handles operator overloading. Skips handling 'get' magic.
8919 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8926 if (SvTHINKFIRST(sv)) {
8927 if (SvREADONLY(sv)) {
8928 Perl_croak_no_modify();
8932 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8934 i = PTR2IV(SvRV(sv));
8938 else sv_force_normal_flags(sv, 0);
8940 flags = SvFLAGS(sv);
8941 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8942 /* It's (privately or publicly) a float, but not tested as an
8943 integer, so test it to see. */
8945 flags = SvFLAGS(sv);
8947 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8948 /* It's publicly an integer, or privately an integer-not-float */
8949 #ifdef PERL_PRESERVE_IVUV
8953 if (SvUVX(sv) == UV_MAX)
8954 sv_setnv(sv, UV_MAX_P1);
8956 (void)SvIOK_only_UV(sv);
8957 SvUV_set(sv, SvUVX(sv) + 1);
8959 if (SvIVX(sv) == IV_MAX)
8960 sv_setuv(sv, (UV)IV_MAX + 1);
8962 (void)SvIOK_only(sv);
8963 SvIV_set(sv, SvIVX(sv) + 1);
8968 if (flags & SVp_NOK) {
8969 const NV was = SvNVX(sv);
8970 if (LIKELY(!Perl_isinfnan(was)) &&
8971 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8972 was >= NV_OVERFLOWS_INTEGERS_AT) {
8973 /* diag_listed_as: Lost precision when %s %f by 1 */
8974 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8975 "Lost precision when incrementing %" NVff " by 1",
8978 (void)SvNOK_only(sv);
8979 SvNV_set(sv, was + 1.0);
8983 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8984 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8985 Perl_croak_no_modify();
8987 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8988 if ((flags & SVTYPEMASK) < SVt_PVIV)
8989 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8990 (void)SvIOK_only(sv);
8995 while (isALPHA(*d)) d++;
8996 while (isDIGIT(*d)) d++;
8997 if (d < SvEND(sv)) {
8998 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8999 #ifdef PERL_PRESERVE_IVUV
9000 /* Got to punt this as an integer if needs be, but we don't issue
9001 warnings. Probably ought to make the sv_iv_please() that does
9002 the conversion if possible, and silently. */
9003 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9004 /* Need to try really hard to see if it's an integer.
9005 9.22337203685478e+18 is an integer.
9006 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9007 so $a="9.22337203685478e+18"; $a+0; $a++
9008 needs to be the same as $a="9.22337203685478e+18"; $a++
9015 /* sv_2iv *should* have made this an NV */
9016 if (flags & SVp_NOK) {
9017 (void)SvNOK_only(sv);
9018 SvNV_set(sv, SvNVX(sv) + 1.0);
9021 /* I don't think we can get here. Maybe I should assert this
9022 And if we do get here I suspect that sv_setnv will croak. NWC
9024 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9025 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9027 #endif /* PERL_PRESERVE_IVUV */
9028 if (!numtype && ckWARN(WARN_NUMERIC))
9029 not_incrementable(sv);
9030 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9034 while (d >= SvPVX_const(sv)) {
9042 /* MKS: The original code here died if letters weren't consecutive.
9043 * at least it didn't have to worry about non-C locales. The
9044 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9045 * arranged in order (although not consecutively) and that only
9046 * [A-Za-z] are accepted by isALPHA in the C locale.
9048 if (isALPHA_FOLD_NE(*d, 'z')) {
9049 do { ++*d; } while (!isALPHA(*d));
9052 *(d--) -= 'z' - 'a';
9057 *(d--) -= 'z' - 'a' + 1;
9061 /* oh,oh, the number grew */
9062 SvGROW(sv, SvCUR(sv) + 2);
9063 SvCUR_set(sv, SvCUR(sv) + 1);
9064 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9075 Auto-decrement of the value in the SV, doing string to numeric conversion
9076 if necessary. Handles 'get' magic and operator overloading.
9082 Perl_sv_dec(pTHX_ SV *const sv)
9091 =for apidoc sv_dec_nomg
9093 Auto-decrement of the value in the SV, doing string to numeric conversion
9094 if necessary. Handles operator overloading. Skips handling 'get' magic.
9100 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9106 if (SvTHINKFIRST(sv)) {
9107 if (SvREADONLY(sv)) {
9108 Perl_croak_no_modify();
9112 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9114 i = PTR2IV(SvRV(sv));
9118 else sv_force_normal_flags(sv, 0);
9120 /* Unlike sv_inc we don't have to worry about string-never-numbers
9121 and keeping them magic. But we mustn't warn on punting */
9122 flags = SvFLAGS(sv);
9123 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9124 /* It's publicly an integer, or privately an integer-not-float */
9125 #ifdef PERL_PRESERVE_IVUV
9129 if (SvUVX(sv) == 0) {
9130 (void)SvIOK_only(sv);
9134 (void)SvIOK_only_UV(sv);
9135 SvUV_set(sv, SvUVX(sv) - 1);
9138 if (SvIVX(sv) == IV_MIN) {
9139 sv_setnv(sv, (NV)IV_MIN);
9143 (void)SvIOK_only(sv);
9144 SvIV_set(sv, SvIVX(sv) - 1);
9149 if (flags & SVp_NOK) {
9152 const NV was = SvNVX(sv);
9153 if (LIKELY(!Perl_isinfnan(was)) &&
9154 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9155 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9156 /* diag_listed_as: Lost precision when %s %f by 1 */
9157 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9158 "Lost precision when decrementing %" NVff " by 1",
9161 (void)SvNOK_only(sv);
9162 SvNV_set(sv, was - 1.0);
9167 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9168 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9169 Perl_croak_no_modify();
9171 if (!(flags & SVp_POK)) {
9172 if ((flags & SVTYPEMASK) < SVt_PVIV)
9173 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9175 (void)SvIOK_only(sv);
9178 #ifdef PERL_PRESERVE_IVUV
9180 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9181 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9182 /* Need to try really hard to see if it's an integer.
9183 9.22337203685478e+18 is an integer.
9184 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9185 so $a="9.22337203685478e+18"; $a+0; $a--
9186 needs to be the same as $a="9.22337203685478e+18"; $a--
9193 /* sv_2iv *should* have made this an NV */
9194 if (flags & SVp_NOK) {
9195 (void)SvNOK_only(sv);
9196 SvNV_set(sv, SvNVX(sv) - 1.0);
9199 /* I don't think we can get here. Maybe I should assert this
9200 And if we do get here I suspect that sv_setnv will croak. NWC
9202 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9203 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9206 #endif /* PERL_PRESERVE_IVUV */
9207 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9210 /* this define is used to eliminate a chunk of duplicated but shared logic
9211 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9212 * used anywhere but here - yves
9214 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9216 SSize_t ix = ++PL_tmps_ix; \
9217 if (UNLIKELY(ix >= PL_tmps_max)) \
9218 ix = tmps_grow_p(ix); \
9219 PL_tmps_stack[ix] = (AnSv); \
9223 =for apidoc sv_mortalcopy
9225 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9226 The new SV is marked as mortal. It will be destroyed "soon", either by an
9227 explicit call to C<FREETMPS>, or by an implicit call at places such as
9228 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9230 =for apidoc sv_mortalcopy_flags
9232 Like C<sv_mortalcopy>, but the extra C<flags> are passed to the
9238 /* Make a string that will exist for the duration of the expression
9239 * evaluation. Actually, it may have to last longer than that, but
9240 * hopefully we won't free it until it has been assigned to a
9241 * permanent location. */
9244 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9248 if (flags & SV_GMAGIC)
9249 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9251 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9252 PUSH_EXTEND_MORTAL__SV_C(sv);
9258 =for apidoc sv_newmortal
9260 Creates a new null SV which is mortal. The reference count of the SV is
9261 set to 1. It will be destroyed "soon", either by an explicit call to
9262 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9263 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9269 Perl_sv_newmortal(pTHX)
9274 SvFLAGS(sv) = SVs_TEMP;
9275 PUSH_EXTEND_MORTAL__SV_C(sv);
9281 =for apidoc newSVpvn_flags
9283 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9284 characters) into it. The reference count for the
9285 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9286 string. You are responsible for ensuring that the source string is at least
9287 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9288 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9289 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9290 returning. If C<SVf_UTF8> is set, C<s>
9291 is considered to be in UTF-8 and the
9292 C<SVf_UTF8> flag will be set on the new SV.
9293 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9295 #define newSVpvn_utf8(s, len, u) \
9296 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9302 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9306 /* All the flags we don't support must be zero.
9307 And we're new code so I'm going to assert this from the start. */
9308 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9310 sv_setpvn(sv,s,len);
9312 /* This code used to do a sv_2mortal(), however we now unroll the call to
9313 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9314 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9315 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9316 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9317 * means that we eliminate quite a few steps than it looks - Yves
9318 * (explaining patch by gfx) */
9320 SvFLAGS(sv) |= flags;
9322 if(flags & SVs_TEMP){
9323 PUSH_EXTEND_MORTAL__SV_C(sv);
9330 =for apidoc sv_2mortal
9332 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9333 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9334 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9335 string buffer can be "stolen" if this SV is copied. See also
9336 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9342 Perl_sv_2mortal(pTHX_ SV *const sv)
9349 PUSH_EXTEND_MORTAL__SV_C(sv);
9357 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9358 characters) into it. The reference count for the
9359 SV is set to 1. If C<len> is zero, Perl will compute the length using
9360 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9361 C<NUL> characters and has to have a terminating C<NUL> byte).
9363 This function can cause reliability issues if you are likely to pass in
9364 empty strings that are not null terminated, because it will run
9365 strlen on the string and potentially run past valid memory.
9367 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9368 For string literals use L</newSVpvs> instead. This function will work fine for
9369 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9370 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9376 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9381 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9386 =for apidoc newSVpvn
9388 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9389 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9390 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9391 are responsible for ensuring that the source buffer is at least
9392 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9399 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9403 sv_setpvn(sv,buffer,len);
9408 =for apidoc newSVhek
9410 Creates a new SV from the hash key structure. It will generate scalars that
9411 point to the shared string table where possible. Returns a new (undefined)
9412 SV if C<hek> is NULL.
9418 Perl_newSVhek(pTHX_ const HEK *const hek)
9427 if (HEK_LEN(hek) == HEf_SVKEY) {
9428 return newSVsv(*(SV**)HEK_KEY(hek));
9430 const int flags = HEK_FLAGS(hek);
9431 if (flags & HVhek_WASUTF8) {
9433 Andreas would like keys he put in as utf8 to come back as utf8
9435 STRLEN utf8_len = HEK_LEN(hek);
9436 SV * const sv = newSV_type(SVt_PV);
9437 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9438 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9439 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9442 } else if (flags & HVhek_UNSHARED) {
9443 /* A hash that isn't using shared hash keys has to have
9444 the flag in every key so that we know not to try to call
9445 share_hek_hek on it. */
9447 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9452 /* This will be overwhelminly the most common case. */
9454 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9455 more efficient than sharepvn(). */
9459 sv_upgrade(sv, SVt_PV);
9460 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9461 SvCUR_set(sv, HEK_LEN(hek));
9473 =for apidoc newSVpvn_share
9475 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9476 table. If the string does not already exist in the table, it is
9477 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9478 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9479 is non-zero, that value is used; otherwise the hash is computed.
9480 The string's hash can later be retrieved from the SV
9481 with the C<SvSHARED_HASH()> macro. The idea here is
9482 that as the string table is used for shared hash keys these strings will have
9483 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9489 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9493 bool is_utf8 = FALSE;
9494 const char *const orig_src = src;
9497 STRLEN tmplen = -len;
9499 /* See the note in hv.c:hv_fetch() --jhi */
9500 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9504 PERL_HASH(hash, src, len);
9506 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9507 changes here, update it there too. */
9508 sv_upgrade(sv, SVt_PV);
9509 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9516 if (src != orig_src)
9522 =for apidoc newSVpv_share
9524 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9531 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9533 return newSVpvn_share(src, strlen(src), hash);
9536 #if defined(PERL_IMPLICIT_CONTEXT)
9538 /* pTHX_ magic can't cope with varargs, so this is a no-context
9539 * version of the main function, (which may itself be aliased to us).
9540 * Don't access this version directly.
9544 Perl_newSVpvf_nocontext(const char *const pat, ...)
9550 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9552 va_start(args, pat);
9553 sv = vnewSVpvf(pat, &args);
9560 =for apidoc newSVpvf
9562 Creates a new SV and initializes it with the string formatted like
9569 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9574 PERL_ARGS_ASSERT_NEWSVPVF;
9576 va_start(args, pat);
9577 sv = vnewSVpvf(pat, &args);
9582 /* backend for newSVpvf() and newSVpvf_nocontext() */
9585 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9589 PERL_ARGS_ASSERT_VNEWSVPVF;
9592 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9599 Creates a new SV and copies a floating point value into it.
9600 The reference count for the SV is set to 1.
9606 Perl_newSVnv(pTHX_ const NV n)
9618 Creates a new SV and copies an integer into it. The reference count for the
9625 Perl_newSViv(pTHX_ const IV i)
9631 /* Inlining ONLY the small relevant subset of sv_setiv here
9632 * for performance. Makes a significant difference. */
9634 /* We're starting from SVt_FIRST, so provided that's
9635 * actual 0, we don't have to unset any SV type flags
9636 * to promote to SVt_IV. */
9637 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9639 SET_SVANY_FOR_BODYLESS_IV(sv);
9640 SvFLAGS(sv) |= SVt_IV;
9652 Creates a new SV and copies an unsigned integer into it.
9653 The reference count for the SV is set to 1.
9659 Perl_newSVuv(pTHX_ const UV u)
9663 /* Inlining ONLY the small relevant subset of sv_setuv here
9664 * for performance. Makes a significant difference. */
9666 /* Using ivs is more efficient than using uvs - see sv_setuv */
9667 if (u <= (UV)IV_MAX) {
9668 return newSViv((IV)u);
9673 /* We're starting from SVt_FIRST, so provided that's
9674 * actual 0, we don't have to unset any SV type flags
9675 * to promote to SVt_IV. */
9676 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9678 SET_SVANY_FOR_BODYLESS_IV(sv);
9679 SvFLAGS(sv) |= SVt_IV;
9681 (void)SvIsUV_on(sv);
9690 =for apidoc newSV_type
9692 Creates a new SV, of the type specified. The reference count for the new SV
9699 Perl_newSV_type(pTHX_ const svtype type)
9704 ASSUME(SvTYPE(sv) == SVt_FIRST);
9705 if(type != SVt_FIRST)
9706 sv_upgrade(sv, type);
9711 =for apidoc newRV_noinc
9713 Creates an RV wrapper for an SV. The reference count for the original
9714 SV is B<not> incremented.
9720 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9724 PERL_ARGS_ASSERT_NEWRV_NOINC;
9728 /* We're starting from SVt_FIRST, so provided that's
9729 * actual 0, we don't have to unset any SV type flags
9730 * to promote to SVt_IV. */
9731 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9733 SET_SVANY_FOR_BODYLESS_IV(sv);
9734 SvFLAGS(sv) |= SVt_IV;
9739 SvRV_set(sv, tmpRef);
9744 /* newRV_inc is the official function name to use now.
9745 * newRV_inc is in fact #defined to newRV in sv.h
9749 Perl_newRV(pTHX_ SV *const sv)
9751 PERL_ARGS_ASSERT_NEWRV;
9753 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9759 Creates a new SV which is an exact duplicate of the original SV.
9762 =for apidoc newSVsv_nomg
9764 Like C<newSVsv> but does not process get magic.
9770 Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags)
9776 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9777 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9780 /* Do this here, otherwise we leak the new SV if this croaks. */
9781 if (flags & SV_GMAGIC)
9784 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC);
9789 =for apidoc sv_reset
9791 Underlying implementation for the C<reset> Perl function.
9792 Note that the perl-level function is vaguely deprecated.
9798 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9800 PERL_ARGS_ASSERT_SV_RESET;
9802 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9806 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9808 char todo[PERL_UCHAR_MAX+1];
9811 if (!stash || SvTYPE(stash) != SVt_PVHV)
9814 if (!s) { /* reset ?? searches */
9815 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9817 const U32 count = mg->mg_len / sizeof(PMOP**);
9818 PMOP **pmp = (PMOP**) mg->mg_ptr;
9819 PMOP *const *const end = pmp + count;
9823 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9825 (*pmp)->op_pmflags &= ~PMf_USED;
9833 /* reset variables */
9835 if (!HvARRAY(stash))
9838 Zero(todo, 256, char);
9842 I32 i = (unsigned char)*s;
9846 max = (unsigned char)*s++;
9847 for ( ; i <= max; i++) {
9850 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9852 for (entry = HvARRAY(stash)[i];
9854 entry = HeNEXT(entry))
9859 if (!todo[(U8)*HeKEY(entry)])
9861 gv = MUTABLE_GV(HeVAL(entry));
9865 if (sv && !SvREADONLY(sv)) {
9866 SV_CHECK_THINKFIRST_COW_DROP(sv);
9867 if (!isGV(sv)) SvOK_off(sv);
9872 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9883 Using various gambits, try to get an IO from an SV: the IO slot if its a
9884 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9885 named after the PV if we're a string.
9887 'Get' magic is ignored on the C<sv> passed in, but will be called on
9888 C<SvRV(sv)> if C<sv> is an RV.
9894 Perl_sv_2io(pTHX_ SV *const sv)
9899 PERL_ARGS_ASSERT_SV_2IO;
9901 switch (SvTYPE(sv)) {
9903 io = MUTABLE_IO(sv);
9907 if (isGV_with_GP(sv)) {
9908 gv = MUTABLE_GV(sv);
9911 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9912 HEKfARG(GvNAME_HEK(gv)));
9918 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9920 SvGETMAGIC(SvRV(sv));
9921 return sv_2io(SvRV(sv));
9923 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9930 if (SvGMAGICAL(sv)) {
9931 newsv = sv_newmortal();
9932 sv_setsv_nomg(newsv, sv);
9934 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9944 Using various gambits, try to get a CV from an SV; in addition, try if
9945 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9946 The flags in C<lref> are passed to C<gv_fetchsv>.
9952 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9957 PERL_ARGS_ASSERT_SV_2CV;
9964 switch (SvTYPE(sv)) {
9968 return MUTABLE_CV(sv);
9978 sv = amagic_deref_call(sv, to_cv_amg);
9981 if (SvTYPE(sv) == SVt_PVCV) {
9982 cv = MUTABLE_CV(sv);
9987 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9988 gv = MUTABLE_GV(sv);
9990 Perl_croak(aTHX_ "Not a subroutine reference");
9992 else if (isGV_with_GP(sv)) {
9993 gv = MUTABLE_GV(sv);
9996 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
10003 /* Some flags to gv_fetchsv mean don't really create the GV */
10004 if (!isGV_with_GP(gv)) {
10008 *st = GvESTASH(gv);
10009 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10010 /* XXX this is probably not what they think they're getting.
10011 * It has the same effect as "sub name;", i.e. just a forward
10020 =for apidoc sv_true
10022 Returns true if the SV has a true value by Perl's rules.
10023 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10024 instead use an in-line version.
10030 Perl_sv_true(pTHX_ SV *const sv)
10035 const XPV* const tXpv = (XPV*)SvANY(sv);
10037 (tXpv->xpv_cur > 1 ||
10038 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10045 return SvIVX(sv) != 0;
10048 return SvNVX(sv) != 0.0;
10050 return sv_2bool(sv);
10056 =for apidoc sv_pvn_force
10058 Get a sensible string out of the SV somehow.
10059 A private implementation of the C<SvPV_force> macro for compilers which
10060 can't cope with complex macro expressions. Always use the macro instead.
10062 =for apidoc sv_pvn_force_flags
10064 Get a sensible string out of the SV somehow.
10065 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10066 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10067 implemented in terms of this function.
10068 You normally want to use the various wrapper macros instead: see
10069 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10075 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10077 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10079 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10080 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10081 sv_force_normal_flags(sv, 0);
10091 if (SvTYPE(sv) > SVt_PVLV
10092 || isGV_with_GP(sv))
10093 /* diag_listed_as: Can't coerce %s to %s in %s */
10094 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10096 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10103 if (SvTYPE(sv) < SVt_PV ||
10104 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10107 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10108 SvGROW(sv, len + 1);
10109 Move(s,SvPVX(sv),len,char);
10110 SvCUR_set(sv, len);
10111 SvPVX(sv)[len] = '\0';
10114 SvPOK_on(sv); /* validate pointer */
10116 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10117 PTR2UV(sv),SvPVX_const(sv)));
10120 (void)SvPOK_only_UTF8(sv);
10121 return SvPVX_mutable(sv);
10125 =for apidoc sv_pvbyten_force
10127 The backend for the C<SvPVbytex_force> macro. Always use the macro
10134 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10136 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10138 sv_pvn_force(sv,lp);
10139 sv_utf8_downgrade(sv,0);
10145 =for apidoc sv_pvutf8n_force
10147 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10154 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10156 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10158 sv_pvn_force(sv,0);
10159 sv_utf8_upgrade_nomg(sv);
10165 =for apidoc sv_reftype
10167 Returns a string describing what the SV is a reference to.
10169 If ob is true and the SV is blessed, the string is the class name,
10170 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10176 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10178 PERL_ARGS_ASSERT_SV_REFTYPE;
10179 if (ob && SvOBJECT(sv)) {
10180 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10183 /* WARNING - There is code, for instance in mg.c, that assumes that
10184 * the only reason that sv_reftype(sv,0) would return a string starting
10185 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10186 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10187 * this routine inside other subs, and it saves time.
10188 * Do not change this assumption without searching for "dodgy type check" in
10191 switch (SvTYPE(sv)) {
10206 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10207 /* tied lvalues should appear to be
10208 * scalars for backwards compatibility */
10209 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10210 ? "SCALAR" : "LVALUE");
10211 case SVt_PVAV: return "ARRAY";
10212 case SVt_PVHV: return "HASH";
10213 case SVt_PVCV: return "CODE";
10214 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10215 ? "GLOB" : "SCALAR");
10216 case SVt_PVFM: return "FORMAT";
10217 case SVt_PVIO: return "IO";
10218 case SVt_INVLIST: return "INVLIST";
10219 case SVt_REGEXP: return "REGEXP";
10220 default: return "UNKNOWN";
10228 Returns a SV describing what the SV passed in is a reference to.
10230 dst can be a SV to be set to the description or NULL, in which case a
10231 mortal SV is returned.
10233 If ob is true and the SV is blessed, the description is the class
10234 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10240 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10242 PERL_ARGS_ASSERT_SV_REF;
10245 dst = sv_newmortal();
10247 if (ob && SvOBJECT(sv)) {
10248 HvNAME_get(SvSTASH(sv))
10249 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10250 : sv_setpvs(dst, "__ANON__");
10253 const char * reftype = sv_reftype(sv, 0);
10254 sv_setpv(dst, reftype);
10260 =for apidoc sv_isobject
10262 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10263 object. If the SV is not an RV, or if the object is not blessed, then this
10270 Perl_sv_isobject(pTHX_ SV *sv)
10286 Returns a boolean indicating whether the SV is blessed into the specified
10287 class. This does not check for subtypes; use C<sv_derived_from> to verify
10288 an inheritance relationship.
10294 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10296 const char *hvname;
10298 PERL_ARGS_ASSERT_SV_ISA;
10308 hvname = HvNAME_get(SvSTASH(sv));
10312 return strEQ(hvname, name);
10316 =for apidoc newSVrv
10318 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10319 RV then it will be upgraded to one. If C<classname> is non-null then the new
10320 SV will be blessed in the specified package. The new SV is returned and its
10321 reference count is 1. The reference count 1 is owned by C<rv>. See also
10322 newRV_inc() and newRV_noinc() for creating a new RV properly.
10328 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10332 PERL_ARGS_ASSERT_NEWSVRV;
10336 SV_CHECK_THINKFIRST_COW_DROP(rv);
10338 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10339 const U32 refcnt = SvREFCNT(rv);
10343 SvREFCNT(rv) = refcnt;
10345 sv_upgrade(rv, SVt_IV);
10346 } else if (SvROK(rv)) {
10347 SvREFCNT_dec(SvRV(rv));
10349 prepare_SV_for_RV(rv);
10357 HV* const stash = gv_stashpv(classname, GV_ADD);
10358 (void)sv_bless(rv, stash);
10364 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10366 SV * const lv = newSV_type(SVt_PVLV);
10367 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10369 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10370 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10371 LvSTARGOFF(lv) = ix;
10372 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10377 =for apidoc sv_setref_pv
10379 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10380 argument will be upgraded to an RV. That RV will be modified to point to
10381 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10382 into the SV. The C<classname> argument indicates the package for the
10383 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10384 will have a reference count of 1, and the RV will be returned.
10386 Do not use with other Perl types such as HV, AV, SV, CV, because those
10387 objects will become corrupted by the pointer copy process.
10389 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10395 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10397 PERL_ARGS_ASSERT_SV_SETREF_PV;
10404 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10409 =for apidoc sv_setref_iv
10411 Copies an 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_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10423 PERL_ARGS_ASSERT_SV_SETREF_IV;
10425 sv_setiv(newSVrv(rv,classname), iv);
10430 =for apidoc sv_setref_uv
10432 Copies an unsigned integer 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_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10444 PERL_ARGS_ASSERT_SV_SETREF_UV;
10446 sv_setuv(newSVrv(rv,classname), uv);
10451 =for apidoc sv_setref_nv
10453 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10454 argument will be upgraded to an RV. That RV will be modified to point to
10455 the new SV. The C<classname> argument indicates the package for the
10456 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10457 will have a reference count of 1, and the RV will be returned.
10463 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10465 PERL_ARGS_ASSERT_SV_SETREF_NV;
10467 sv_setnv(newSVrv(rv,classname), nv);
10472 =for apidoc sv_setref_pvn
10474 Copies a string into a new SV, optionally blessing the SV. The length of the
10475 string must be specified with C<n>. The C<rv> argument will be upgraded to
10476 an RV. That RV will be modified to point to the new SV. The C<classname>
10477 argument indicates the package for the blessing. Set C<classname> to
10478 C<NULL> to avoid the blessing. The new SV will have a reference count
10479 of 1, and the RV will be returned.
10481 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10487 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10488 const char *const pv, const STRLEN n)
10490 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10492 sv_setpvn(newSVrv(rv,classname), pv, n);
10497 =for apidoc sv_bless
10499 Blesses an SV into a specified package. The SV must be an RV. The package
10500 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10501 of the SV is unaffected.
10507 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10510 HV *oldstash = NULL;
10512 PERL_ARGS_ASSERT_SV_BLESS;
10516 Perl_croak(aTHX_ "Can't bless non-reference value");
10518 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10519 if (SvREADONLY(tmpRef))
10520 Perl_croak_no_modify();
10521 if (SvOBJECT(tmpRef)) {
10522 oldstash = SvSTASH(tmpRef);
10525 SvOBJECT_on(tmpRef);
10526 SvUPGRADE(tmpRef, SVt_PVMG);
10527 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10528 SvREFCNT_dec(oldstash);
10530 if(SvSMAGICAL(tmpRef))
10531 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10539 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10540 * as it is after unglobbing it.
10543 PERL_STATIC_INLINE void
10544 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10548 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10550 PERL_ARGS_ASSERT_SV_UNGLOB;
10552 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10554 if (!(flags & SV_COW_DROP_PV))
10555 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10557 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10559 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10560 && HvNAME_get(stash))
10561 mro_method_changed_in(stash);
10562 gp_free(MUTABLE_GV(sv));
10565 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10566 GvSTASH(sv) = NULL;
10569 if (GvNAME_HEK(sv)) {
10570 unshare_hek(GvNAME_HEK(sv));
10572 isGV_with_GP_off(sv);
10574 if(SvTYPE(sv) == SVt_PVGV) {
10575 /* need to keep SvANY(sv) in the right arena */
10576 xpvmg = new_XPVMG();
10577 StructCopy(SvANY(sv), xpvmg, XPVMG);
10578 del_XPVGV(SvANY(sv));
10581 SvFLAGS(sv) &= ~SVTYPEMASK;
10582 SvFLAGS(sv) |= SVt_PVMG;
10585 /* Intentionally not calling any local SET magic, as this isn't so much a
10586 set operation as merely an internal storage change. */
10587 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10588 else sv_setsv_flags(sv, temp, 0);
10590 if ((const GV *)sv == PL_last_in_gv)
10591 PL_last_in_gv = NULL;
10592 else if ((const GV *)sv == PL_statgv)
10597 =for apidoc sv_unref_flags
10599 Unsets the RV status of the SV, and decrements the reference count of
10600 whatever was being referenced by the RV. This can almost be thought of
10601 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10602 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10603 (otherwise the decrementing is conditional on the reference count being
10604 different from one or the reference being a readonly SV).
10605 See C<L</SvROK_off>>.
10611 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10613 SV* const target = SvRV(ref);
10615 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10617 if (SvWEAKREF(ref)) {
10618 sv_del_backref(target, ref);
10619 SvWEAKREF_off(ref);
10620 SvRV_set(ref, NULL);
10623 SvRV_set(ref, NULL);
10625 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10626 assigned to as BEGIN {$a = \"Foo"} will fail. */
10627 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10628 SvREFCNT_dec_NN(target);
10629 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10630 sv_2mortal(target); /* Schedule for freeing later */
10634 =for apidoc sv_untaint
10636 Untaint an SV. Use C<SvTAINTED_off> instead.
10642 Perl_sv_untaint(pTHX_ SV *const sv)
10644 PERL_ARGS_ASSERT_SV_UNTAINT;
10645 PERL_UNUSED_CONTEXT;
10647 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10648 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10655 =for apidoc sv_tainted
10657 Test an SV for taintedness. Use C<SvTAINTED> instead.
10663 Perl_sv_tainted(pTHX_ SV *const sv)
10665 PERL_ARGS_ASSERT_SV_TAINTED;
10666 PERL_UNUSED_CONTEXT;
10668 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10669 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10670 if (mg && (mg->mg_len & 1) )
10676 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10677 private to this file */
10680 =for apidoc sv_setpviv
10682 Copies an integer into the given SV, also updating its string value.
10683 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10689 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10691 /* The purpose of this union is to ensure that arr is aligned on
10692 a 2 byte boundary, because that is what uiv_2buf() requires */
10694 char arr[TYPE_CHARS(UV)];
10698 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10700 PERL_ARGS_ASSERT_SV_SETPVIV;
10702 sv_setpvn(sv, ptr, ebuf - ptr);
10706 =for apidoc sv_setpviv_mg
10708 Like C<sv_setpviv>, but also handles 'set' magic.
10714 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10716 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10718 sv_setpviv(sv, iv);
10722 #endif /* NO_MATHOMS */
10724 #if defined(PERL_IMPLICIT_CONTEXT)
10726 /* pTHX_ magic can't cope with varargs, so this is a no-context
10727 * version of the main function, (which may itself be aliased to us).
10728 * Don't access this version directly.
10732 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10737 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10739 va_start(args, pat);
10740 sv_vsetpvf(sv, pat, &args);
10744 /* pTHX_ magic can't cope with varargs, so this is a no-context
10745 * version of the main function, (which may itself be aliased to us).
10746 * Don't access this version directly.
10750 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10755 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10757 va_start(args, pat);
10758 sv_vsetpvf_mg(sv, pat, &args);
10764 =for apidoc sv_setpvf
10766 Works like C<sv_catpvf> but copies the text into the SV instead of
10767 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10773 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10777 PERL_ARGS_ASSERT_SV_SETPVF;
10779 va_start(args, pat);
10780 sv_vsetpvf(sv, pat, &args);
10785 =for apidoc sv_vsetpvf
10787 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10788 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10790 Usually used via its frontend C<sv_setpvf>.
10796 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10798 PERL_ARGS_ASSERT_SV_VSETPVF;
10800 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10804 =for apidoc sv_setpvf_mg
10806 Like C<sv_setpvf>, but also handles 'set' magic.
10812 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10816 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10818 va_start(args, pat);
10819 sv_vsetpvf_mg(sv, pat, &args);
10824 =for apidoc sv_vsetpvf_mg
10826 Like C<sv_vsetpvf>, but also handles 'set' magic.
10828 Usually used via its frontend C<sv_setpvf_mg>.
10834 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10836 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10838 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10842 #if defined(PERL_IMPLICIT_CONTEXT)
10844 /* pTHX_ magic can't cope with varargs, so this is a no-context
10845 * version of the main function, (which may itself be aliased to us).
10846 * Don't access this version directly.
10850 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10855 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10857 va_start(args, pat);
10858 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10862 /* pTHX_ magic can't cope with varargs, so this is a no-context
10863 * version of the main function, (which may itself be aliased to us).
10864 * Don't access this version directly.
10868 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10873 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10875 va_start(args, pat);
10876 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10883 =for apidoc sv_catpvf
10885 Processes its arguments like C<sprintf>, and appends the formatted
10886 output to an SV. As with C<sv_vcatpvfn> called with a non-null C-style
10887 variable argument list, argument reordering is not supported.
10888 If the appended data contains "wide" characters
10889 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10890 and characters >255 formatted with C<%c>), the original SV might get
10891 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10892 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10893 valid UTF-8; if the original SV was bytes, the pattern should be too.
10898 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10902 PERL_ARGS_ASSERT_SV_CATPVF;
10904 va_start(args, pat);
10905 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10910 =for apidoc sv_vcatpvf
10912 Processes its arguments like C<sv_vcatpvfn> called with a non-null C-style
10913 variable argument list, and appends the formatted output
10914 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10916 Usually used via its frontend C<sv_catpvf>.
10922 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10924 PERL_ARGS_ASSERT_SV_VCATPVF;
10926 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10930 =for apidoc sv_catpvf_mg
10932 Like C<sv_catpvf>, but also handles 'set' magic.
10938 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10942 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10944 va_start(args, pat);
10945 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10951 =for apidoc sv_vcatpvf_mg
10953 Like C<sv_vcatpvf>, but also handles 'set' magic.
10955 Usually used via its frontend C<sv_catpvf_mg>.
10961 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10963 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10965 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10970 =for apidoc sv_vsetpvfn
10972 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10975 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10981 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10982 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
10984 PERL_ARGS_ASSERT_SV_VSETPVFN;
10987 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
10991 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
10993 PERL_STATIC_INLINE void
10994 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
10996 STRLEN const need = len + SvCUR(sv) + 1;
10999 /* can't wrap as both len and SvCUR() are allocated in
11000 * memory and together can't consume all the address space
11002 assert(need > len);
11007 Copy(buf, end, len, char);
11010 SvCUR_set(sv, need - 1);
11015 * Warn of missing argument to sprintf. The value used in place of such
11016 * arguments should be &PL_sv_no; an undefined value would yield
11017 * inappropriate "use of uninit" warnings [perl #71000].
11020 S_warn_vcatpvfn_missing_argument(pTHX) {
11021 if (ckWARN(WARN_MISSING)) {
11022 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11023 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11032 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11033 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11037 /* Given an int i from the next arg (if args is true) or an sv from an arg
11038 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11039 * with overflow checking.
11040 * Sets *neg to true if the value was negative (untouched otherwise.
11041 * Returns the absolute value.
11042 * As an extra margin of safety, it croaks if the returned value would
11043 * exceed the maximum value of a STRLEN / 4.
11047 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11061 if (UNLIKELY(SvIsUV(sv))) {
11062 UV uv = SvUV_nomg(sv);
11064 S_croak_overflow();
11068 iv = SvIV_nomg(sv);
11072 S_croak_overflow();
11078 if (iv > (IV)(((STRLEN)~0) / 4))
11079 S_croak_overflow();
11084 /* Read in and return a number. Updates *pattern to point to the char
11085 * following the number. Expects the first char to 1..9.
11086 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11087 * This is a belt-and-braces safety measure to complement any
11088 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11089 * It means that e.g. on a 32-bit system the width/precision can't be more
11090 * than 1G, which seems reasonable.
11094 S_expect_number(pTHX_ const char **const pattern)
11098 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11100 assert(inRANGE(**pattern, '1', '9'));
11102 var = *(*pattern)++ - '0';
11103 while (isDIGIT(**pattern)) {
11104 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11105 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11106 S_croak_overflow();
11107 var = var * 10 + (*(*pattern)++ - '0');
11112 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11113 * ensures it's big enough), back fill it with the rounded integer part of
11114 * nv. Returns ptr to start of string, and sets *len to its length.
11115 * Returns NULL if not convertible.
11119 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11121 const int neg = nv < 0;
11124 PERL_ARGS_ASSERT_F0CONVERT;
11126 assert(!Perl_isinfnan(nv));
11129 if (nv != 0.0 && nv < UV_MAX) {
11135 if (uv & 1 && uv == nv)
11136 uv--; /* Round to even */
11139 const unsigned dig = uv % 10;
11141 } while (uv /= 10);
11151 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11154 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11155 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11157 PERL_ARGS_ASSERT_SV_VCATPVFN;
11159 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11163 /* For the vcatpvfn code, we need a long double target in case
11164 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11165 * with long double formats, even without NV being long double. But we
11166 * call the target 'fv' instead of 'nv', since most of the time it is not
11167 * (most compilers these days recognize "long double", even if only as a
11168 * synonym for "double").
11170 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11171 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11172 # define VCATPVFN_FV_GF PERL_PRIgldbl
11173 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11174 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11175 # define VCATPVFN_NV_TO_FV(nv,fv) \
11178 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11181 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11183 typedef long double vcatpvfn_long_double_t;
11185 # define VCATPVFN_FV_GF NVgf
11186 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11187 typedef NV vcatpvfn_long_double_t;
11190 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11191 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11192 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11193 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11194 * after the first 1023 zero bits.
11196 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11197 * of dynamically growing buffer might be better, start at just 16 bytes
11198 * (for example) and grow only when necessary. Or maybe just by looking
11199 * at the exponents of the two doubles? */
11200 # define DOUBLEDOUBLE_MAXBITS 2098
11203 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11204 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11205 * per xdigit. For the double-double case, this can be rather many.
11206 * The non-double-double-long-double overshoots since all bits of NV
11207 * are not mantissa bits, there are also exponent bits. */
11208 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11209 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11211 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11214 /* If we do not have a known long double format, (including not using
11215 * long doubles, or long doubles being equal to doubles) then we will
11216 * fall back to the ldexp/frexp route, with which we can retrieve at
11217 * most as many bits as our widest unsigned integer type is. We try
11218 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11220 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11221 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11223 #if defined(HAS_QUAD) && defined(Uquad_t)
11224 # define MANTISSATYPE Uquad_t
11225 # define MANTISSASIZE 8
11227 # define MANTISSATYPE UV
11228 # define MANTISSASIZE UVSIZE
11231 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11232 # define HEXTRACT_LITTLE_ENDIAN
11233 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11234 # define HEXTRACT_BIG_ENDIAN
11236 # define HEXTRACT_MIX_ENDIAN
11239 /* S_hextract() is a helper for S_format_hexfp, for extracting
11240 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11241 * are being extracted from (either directly from the long double in-memory
11242 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11243 * is used to update the exponent. The subnormal is set to true
11244 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11245 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11247 * The tricky part is that S_hextract() needs to be called twice:
11248 * the first time with vend as NULL, and the second time with vend as
11249 * the pointer returned by the first call. What happens is that on
11250 * the first round the output size is computed, and the intended
11251 * extraction sanity checked. On the second round the actual output
11252 * (the extraction of the hexadecimal values) takes place.
11253 * Sanity failures cause fatal failures during both rounds. */
11255 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11256 U8* vhex, U8* vend)
11260 int ixmin = 0, ixmax = 0;
11262 /* XXX Inf/NaN are not handled here, since it is
11263 * assumed they are to be output as "Inf" and "NaN". */
11265 /* These macros are just to reduce typos, they have multiple
11266 * repetitions below, but usually only one (or sometimes two)
11267 * of them is really being used. */
11268 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11269 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11270 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11271 #define HEXTRACT_OUTPUT(ix) \
11273 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11275 #define HEXTRACT_COUNT(ix, c) \
11277 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11279 #define HEXTRACT_BYTE(ix) \
11281 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11283 #define HEXTRACT_LO_NYBBLE(ix) \
11285 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11287 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11288 * to make it look less odd when the top bits of a NV
11289 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11290 * order bits can be in the "low nybble" of a byte. */
11291 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11292 #define HEXTRACT_BYTES_LE(a, b) \
11293 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11294 #define HEXTRACT_BYTES_BE(a, b) \
11295 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11296 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11297 #define HEXTRACT_IMPLICIT_BIT(nv) \
11299 if (!*subnormal) { \
11300 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11304 /* Most formats do. Those which don't should undef this.
11306 * But also note that IEEE 754 subnormals do not have it, or,
11307 * expressed alternatively, their implicit bit is zero. */
11308 #define HEXTRACT_HAS_IMPLICIT_BIT
11310 /* Many formats do. Those which don't should undef this. */
11311 #define HEXTRACT_HAS_TOP_NYBBLE
11313 /* HEXTRACTSIZE is the maximum number of xdigits. */
11314 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11315 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11317 # define HEXTRACTSIZE 2 * NVSIZE
11320 const U8* vmaxend = vhex + HEXTRACTSIZE;
11322 assert(HEXTRACTSIZE <= VHEX_SIZE);
11324 PERL_UNUSED_VAR(ix); /* might happen */
11325 (void)Perl_frexp(PERL_ABS(nv), exponent);
11326 *subnormal = FALSE;
11327 if (vend && (vend <= vhex || vend > vmaxend)) {
11328 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11329 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11332 /* First check if using long doubles. */
11333 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11334 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11335 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11336 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11337 /* The bytes 13..0 are the mantissa/fraction,
11338 * the 15,14 are the sign+exponent. */
11339 const U8* nvp = (const U8*)(&nv);
11340 HEXTRACT_GET_SUBNORMAL(nv);
11341 HEXTRACT_IMPLICIT_BIT(nv);
11342 # undef HEXTRACT_HAS_TOP_NYBBLE
11343 HEXTRACT_BYTES_LE(13, 0);
11344 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11345 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11346 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11347 /* The bytes 2..15 are the mantissa/fraction,
11348 * the 0,1 are the sign+exponent. */
11349 const U8* nvp = (const U8*)(&nv);
11350 HEXTRACT_GET_SUBNORMAL(nv);
11351 HEXTRACT_IMPLICIT_BIT(nv);
11352 # undef HEXTRACT_HAS_TOP_NYBBLE
11353 HEXTRACT_BYTES_BE(2, 15);
11354 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11355 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11356 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11357 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11358 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11359 /* The bytes 0..1 are the sign+exponent,
11360 * the bytes 2..9 are the mantissa/fraction. */
11361 const U8* nvp = (const U8*)(&nv);
11362 # undef HEXTRACT_HAS_IMPLICIT_BIT
11363 # undef HEXTRACT_HAS_TOP_NYBBLE
11364 HEXTRACT_GET_SUBNORMAL(nv);
11365 HEXTRACT_BYTES_LE(7, 0);
11366 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11367 /* Does this format ever happen? (Wikipedia says the Motorola
11368 * 6888x math coprocessors used format _like_ this but padded
11369 * to 96 bits with 16 unused bits between the exponent and the
11371 const U8* nvp = (const U8*)(&nv);
11372 # undef HEXTRACT_HAS_IMPLICIT_BIT
11373 # undef HEXTRACT_HAS_TOP_NYBBLE
11374 HEXTRACT_GET_SUBNORMAL(nv);
11375 HEXTRACT_BYTES_BE(0, 7);
11377 # define HEXTRACT_FALLBACK
11378 /* Double-double format: two doubles next to each other.
11379 * The first double is the high-order one, exactly like
11380 * it would be for a "lone" double. The second double
11381 * is shifted down using the exponent so that that there
11382 * are no common bits. The tricky part is that the value
11383 * of the double-double is the SUM of the two doubles and
11384 * the second one can be also NEGATIVE.
11386 * Because of this tricky construction the bytewise extraction we
11387 * use for the other long double formats doesn't work, we must
11388 * extract the values bit by bit.
11390 * The little-endian double-double is used .. somewhere?
11392 * The big endian double-double is used in e.g. PPC/Power (AIX)
11395 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11396 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11397 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11400 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11401 /* Using normal doubles, not long doubles.
11403 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11404 * bytes, since we might need to handle printf precision, and
11405 * also need to insert the radix. */
11407 # ifdef HEXTRACT_LITTLE_ENDIAN
11408 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11409 const U8* nvp = (const U8*)(&nv);
11410 HEXTRACT_GET_SUBNORMAL(nv);
11411 HEXTRACT_IMPLICIT_BIT(nv);
11412 HEXTRACT_TOP_NYBBLE(6);
11413 HEXTRACT_BYTES_LE(5, 0);
11414 # elif defined(HEXTRACT_BIG_ENDIAN)
11415 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11416 const U8* nvp = (const U8*)(&nv);
11417 HEXTRACT_GET_SUBNORMAL(nv);
11418 HEXTRACT_IMPLICIT_BIT(nv);
11419 HEXTRACT_TOP_NYBBLE(1);
11420 HEXTRACT_BYTES_BE(2, 7);
11421 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11422 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11423 const U8* nvp = (const U8*)(&nv);
11424 HEXTRACT_GET_SUBNORMAL(nv);
11425 HEXTRACT_IMPLICIT_BIT(nv);
11426 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11427 HEXTRACT_BYTE(1); /* 5 */
11428 HEXTRACT_BYTE(0); /* 4 */
11429 HEXTRACT_BYTE(7); /* 3 */
11430 HEXTRACT_BYTE(6); /* 2 */
11431 HEXTRACT_BYTE(5); /* 1 */
11432 HEXTRACT_BYTE(4); /* 0 */
11433 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11434 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11435 const U8* nvp = (const U8*)(&nv);
11436 HEXTRACT_GET_SUBNORMAL(nv);
11437 HEXTRACT_IMPLICIT_BIT(nv);
11438 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11439 HEXTRACT_BYTE(6); /* 5 */
11440 HEXTRACT_BYTE(7); /* 4 */
11441 HEXTRACT_BYTE(0); /* 3 */
11442 HEXTRACT_BYTE(1); /* 2 */
11443 HEXTRACT_BYTE(2); /* 1 */
11444 HEXTRACT_BYTE(3); /* 0 */
11446 # define HEXTRACT_FALLBACK
11449 # define HEXTRACT_FALLBACK
11451 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11453 #ifdef HEXTRACT_FALLBACK
11454 HEXTRACT_GET_SUBNORMAL(nv);
11455 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11456 /* The fallback is used for the double-double format, and
11457 * for unknown long double formats, and for unknown double
11458 * formats, or in general unknown NV formats. */
11459 if (nv == (NV)0.0) {
11467 NV d = nv < 0 ? -nv : nv;
11469 U8 ha = 0x0; /* hexvalue accumulator */
11470 U8 hd = 0x8; /* hexvalue digit */
11472 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11473 * this is essentially manual frexp(). Multiplying by 0.5 and
11474 * doubling should be lossless in binary floating point. */
11484 while (d >= e + e) {
11488 /* Now e <= d < 2*e */
11490 /* First extract the leading hexdigit (the implicit bit). */
11506 /* Then extract the remaining hexdigits. */
11507 while (d > (NV)0.0) {
11513 /* Output or count in groups of four bits,
11514 * that is, when the hexdigit is down to one. */
11519 /* Reset the hexvalue. */
11528 /* Flush possible pending hexvalue. */
11538 /* Croak for various reasons: if the output pointer escaped the
11539 * output buffer, if the extraction index escaped the extraction
11540 * buffer, or if the ending output pointer didn't match the
11541 * previously computed value. */
11542 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11543 /* For double-double the ixmin and ixmax stay at zero,
11544 * which is convenient since the HEXTRACTSIZE is tricky
11545 * for double-double. */
11546 ixmin < 0 || ixmax >= NVSIZE ||
11547 (vend && v != vend)) {
11548 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11549 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11555 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11557 * Processes the %a/%A hexadecimal floating-point format, since the
11558 * built-in snprintf()s which are used for most of the f/p formats, don't
11559 * universally handle %a/%A.
11560 * Populates buf of length bufsize, and returns the length of the created
11562 * The rest of the args have the same meaning as the local vars of the
11563 * same name within Perl_sv_vcatpvfn_flags().
11565 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11567 * It requires the caller to make buf large enough.
11571 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11572 const NV nv, const vcatpvfn_long_double_t fv,
11573 bool has_precis, STRLEN precis, STRLEN width,
11574 bool alt, char plus, bool left, bool fill)
11576 /* Hexadecimal floating point. */
11578 U8 vhex[VHEX_SIZE];
11579 U8* v = vhex; /* working pointer to vhex */
11580 U8* vend; /* pointer to one beyond last digit of vhex */
11581 U8* vfnz = NULL; /* first non-zero */
11582 U8* vlnz = NULL; /* last non-zero */
11583 U8* v0 = NULL; /* first output */
11584 const bool lower = (c == 'a');
11585 /* At output the values of vhex (up to vend) will
11586 * be mapped through the xdig to get the actual
11587 * human-readable xdigits. */
11588 const char* xdig = PL_hexdigit;
11589 STRLEN zerotail = 0; /* how many extra zeros to append */
11590 int exponent = 0; /* exponent of the floating point input */
11591 bool hexradix = FALSE; /* should we output the radix */
11592 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11593 bool negative = FALSE;
11596 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11598 * For example with denormals, (assuming the vanilla
11599 * 64-bit double): the exponent is zero. 1xp-1074 is
11600 * the smallest denormal and the smallest double, it
11601 * could be output also as 0x0.0000000000001p-1022 to
11602 * match its internal structure. */
11604 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11605 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11607 #if NVSIZE > DOUBLESIZE
11608 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11609 /* In this case there is an implicit bit,
11610 * and therefore the exponent is shifted by one. */
11612 # elif defined(NV_X86_80_BIT)
11614 /* The subnormals of the x86-80 have a base exponent of -16382,
11615 * (while the physical exponent bits are zero) but the frexp()
11616 * returned the scientific-style floating exponent. We want
11617 * to map the last one as:
11618 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11619 * -16835..-16388 -> -16384
11620 * since we want to keep the first hexdigit
11621 * as one of the [8421]. */
11622 exponent = -4 * ( (exponent + 1) / -4) - 2;
11626 /* TBD: other non-implicit-bit platforms than the x86-80. */
11630 negative = fv < 0 || Perl_signbit(nv);
11641 xdig += 16; /* Use uppercase hex. */
11644 /* Find the first non-zero xdigit. */
11645 for (v = vhex; v < vend; v++) {
11653 /* Find the last non-zero xdigit. */
11654 for (v = vend - 1; v >= vhex; v--) {
11661 #if NVSIZE == DOUBLESIZE
11667 #ifndef NV_X86_80_BIT
11669 /* IEEE 754 subnormals (but not the x86 80-bit):
11670 * we want "normalize" the subnormal,
11671 * so we need to right shift the hex nybbles
11672 * so that the output of the subnormal starts
11673 * from the first true bit. (Another, equally
11674 * valid, policy would be to dump the subnormal
11675 * nybbles as-is, to display the "physical" layout.) */
11678 /* Find the ceil(log2(v[0])) of
11679 * the top non-zero nybble. */
11680 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11684 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11685 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11699 U8* ve = (subnormal ? vlnz + 1 : vend);
11700 SSize_t vn = ve - v0;
11702 if (precis < (Size_t)(vn - 1)) {
11703 bool overflow = FALSE;
11704 if (v0[precis + 1] < 0x8) {
11705 /* Round down, nothing to do. */
11706 } else if (v0[precis + 1] > 0x8) {
11709 overflow = v0[precis] > 0xF;
11711 } else { /* v0[precis] == 0x8 */
11712 /* Half-point: round towards the one
11713 * with the even least-significant digit:
11721 * 78 -> 8 f8 -> 10 */
11722 if ((v0[precis] & 0x1)) {
11725 overflow = v0[precis] > 0xF;
11730 for (v = v0 + precis - 1; v >= v0; v--) {
11732 overflow = *v > 0xF;
11738 if (v == v0 - 1 && overflow) {
11739 /* If the overflow goes all the
11740 * way to the front, we need to
11741 * insert 0x1 in front, and adjust
11743 Move(v0, v0 + 1, vn - 1, char);
11749 /* The new effective "last non zero". */
11750 vlnz = v0 + precis;
11754 subnormal ? precis - vn + 1 :
11755 precis - (vlnz - vhex);
11762 /* If there are non-zero xdigits, the radix
11763 * is output after the first one. */
11771 zerotail = has_precis ? precis : 0;
11774 /* The radix is always output if precis, or if alt. */
11775 if ((has_precis && precis > 0) || alt) {
11780 #ifndef USE_LOCALE_NUMERIC
11783 if (IN_LC(LC_NUMERIC)) {
11785 const char* r = SvPV(PL_numeric_radix_sv, n);
11786 Copy(r, p, n, char);
11800 if (zerotail > 0) {
11801 while (zerotail--) {
11808 /* sanity checks */
11809 if (elen >= bufsize || width >= bufsize)
11810 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11811 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11813 elen += my_snprintf(p, bufsize - elen,
11814 "%c%+d", lower ? 'p' : 'P',
11817 if (elen < width) {
11818 STRLEN gap = (STRLEN)(width - elen);
11820 /* Pad the back with spaces. */
11821 memset(buf + elen, ' ', gap);
11824 /* Insert the zeros after the "0x" and the
11825 * the potential sign, but before the digits,
11826 * otherwise we end up with "0000xH.HHH...",
11827 * when we want "0x000H.HHH..." */
11828 STRLEN nzero = gap;
11829 char* zerox = buf + 2;
11830 STRLEN nmove = elen - 2;
11831 if (negative || plus) {
11835 Move(zerox, zerox + nzero, nmove, char);
11836 memset(zerox, fill ? '0' : ' ', nzero);
11839 /* Move it to the right. */
11840 Move(buf, buf + gap,
11842 /* Pad the front with spaces. */
11843 memset(buf, ' ', gap);
11852 =for apidoc sv_vcatpvfn
11854 =for apidoc sv_vcatpvfn_flags
11856 Processes its arguments like C<vsprintf> and appends the formatted output
11857 to an SV. Uses an array of SVs if the C-style variable argument list is
11858 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11859 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11860 C<va_list> argument list with a format string that uses argument reordering
11861 will yield an exception.
11863 When running with taint checks enabled, indicates via
11864 C<maybe_tainted> if results are untrustworthy (often due to the use of
11867 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11869 It assumes that pat has the same utf8-ness as sv. It's the caller's
11870 responsibility to ensure that this is so.
11872 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11879 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11880 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11883 const char *fmtstart; /* character following the current '%' */
11884 const char *q; /* current position within format */
11885 const char *patend;
11888 static const char nullstr[] = "(null)";
11889 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11890 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11891 /* Times 4: a decimal digit takes more than 3 binary digits.
11892 * NV_DIG: mantissa takes that many decimal digits.
11893 * Plus 32: Playing safe. */
11894 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11895 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11896 #ifdef USE_LOCALE_NUMERIC
11897 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11898 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11901 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11902 PERL_UNUSED_ARG(maybe_tainted);
11904 if (flags & SV_GMAGIC)
11907 /* no matter what, this is a string now */
11908 (void)SvPV_force_nomg(sv, origlen);
11910 /* the code that scans for flags etc following a % relies on
11911 * a '\0' being present to avoid falling off the end. Ideally that
11912 * should be fixed */
11913 assert(pat[patlen] == '\0');
11916 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11917 * In each case, if there isn't the correct number of args, instead
11918 * fall through to the main code to handle the issuing of any
11922 if (patlen == 0 && (args || sv_count == 0))
11925 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11928 if (patlen == 2 && pat[1] == 's') {
11930 const char * const s = va_arg(*args, char*);
11931 sv_catpv_nomg(sv, s ? s : nullstr);
11934 /* we want get magic on the source but not the target.
11935 * sv_catsv can't do that, though */
11936 SvGETMAGIC(*svargs);
11937 sv_catsv_nomg(sv, *svargs);
11944 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11945 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11946 sv_catsv_nomg(sv, asv);
11950 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11951 /* special-case "%.0f" */
11952 else if ( patlen == 4
11953 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11955 const NV nv = SvNV(*svargs);
11956 if (LIKELY(!Perl_isinfnan(nv))) {
11960 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11961 sv_catpvn_nomg(sv, p, l);
11966 #endif /* !USE_LONG_DOUBLE */
11970 patend = (char*)pat + patlen;
11971 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
11972 char intsize = 0; /* size qualifier in "%hi..." etc */
11973 bool alt = FALSE; /* has "%#..." */
11974 bool left = FALSE; /* has "%-..." */
11975 bool fill = FALSE; /* has "%0..." */
11976 char plus = 0; /* has "%+..." */
11977 STRLEN width = 0; /* value of "%NNN..." */
11978 bool has_precis = FALSE; /* has "%.NNN..." */
11979 STRLEN precis = 0; /* value of "%.NNN..." */
11980 int base = 0; /* base to print in, e.g. 8 for %o */
11981 UV uv = 0; /* the value to print of int-ish args */
11983 bool vectorize = FALSE; /* has "%v..." */
11984 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
11985 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
11986 STRLEN veclen = 0; /* SvCUR(vec arg) */
11987 const char *dotstr = NULL; /* separator string for %v */
11988 STRLEN dotstrlen; /* length of separator string for %v */
11990 Size_t efix = 0; /* explicit format parameter index */
11991 const Size_t osvix = svix; /* original index in case of bad fmt */
11994 bool is_utf8 = FALSE; /* is this item utf8? */
11995 bool arg_missing = FALSE; /* give "Missing argument" warning */
11996 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
11997 STRLEN esignlen = 0; /* length of e.g. "-0x" */
11998 STRLEN zeros = 0; /* how many '0' to prepend */
12000 const char *eptr = NULL; /* the address of the element string */
12001 STRLEN elen = 0; /* the length of the element string */
12003 char c; /* the actual format ('d', s' etc) */
12006 /* echo everything up to the next format specification */
12007 for (q = fmtstart; q < patend && *q != '%'; ++q)
12010 if (q > fmtstart) {
12011 if (has_utf8 && !pat_utf8) {
12012 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12016 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12018 for (p = fmtstart; p < q; p++)
12019 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12024 for (p = fmtstart; p < q; p++)
12025 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12027 SvCUR_set(sv, need - 1);
12030 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12035 fmtstart = q; /* fmtstart is char following the '%' */
12038 We allow format specification elements in this order:
12039 \d+\$ explicit format parameter index
12041 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12042 0 flag (as above): repeated to allow "v02"
12043 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12044 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12046 [%bcdefginopsuxDFOUX] format (mandatory)
12049 if (inRANGE(*q, '1', '9')) {
12050 width = expect_number(&q);
12053 Perl_croak_nocontext(
12054 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12056 efix = (Size_t)width;
12058 no_redundant_warning = TRUE;
12070 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12097 /* at this point we can expect one of:
12099 * 123 an explicit width
12100 * * width taken from next arg
12101 * *12$ width taken from 12th arg
12104 * But any width specification may be preceded by a v, in one of its
12109 * So an asterisk may be either a width specifier or a vector
12110 * separator arg specifier, and we don't know which initially
12115 STRLEN ix; /* explicit width/vector separator index */
12117 if (inRANGE(*q, '1', '9')) {
12118 ix = expect_number(&q);
12121 Perl_croak_nocontext(
12122 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12123 no_redundant_warning = TRUE;
12132 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12133 * with the default "." */
12138 vecsv = va_arg(*args, SV*);
12140 ix = ix ? ix - 1 : svix++;
12141 vecsv = ix < sv_count ? svargs[ix]
12142 : (arg_missing = TRUE, &PL_sv_no);
12144 dotstr = SvPV_const(vecsv, dotstrlen);
12145 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12146 bad with tied or overloaded values that return UTF8. */
12147 if (DO_UTF8(vecsv))
12149 else if (has_utf8) {
12150 vecsv = sv_mortalcopy(vecsv);
12151 sv_utf8_upgrade(vecsv);
12152 dotstr = SvPV_const(vecsv, dotstrlen);
12159 /* the asterisk specified a width */
12164 i = va_arg(*args, int);
12166 ix = ix ? ix - 1 : svix++;
12167 sv = (ix < sv_count) ? svargs[ix]
12168 : (arg_missing = TRUE, (SV*)NULL);
12170 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12173 else if (*q == 'v') {
12184 /* explicit width? */
12189 if (inRANGE(*q, '1', '9'))
12190 width = expect_number(&q);
12200 STRLEN ix; /* explicit precision index */
12202 if (inRANGE(*q, '1', '9')) {
12203 ix = expect_number(&q);
12206 Perl_croak_nocontext(
12207 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12208 no_redundant_warning = TRUE;
12221 i = va_arg(*args, int);
12223 ix = ix ? ix - 1 : svix++;
12224 sv = (ix < sv_count) ? svargs[ix]
12225 : (arg_missing = TRUE, (SV*)NULL);
12227 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12229 /* ignore negative precision */
12235 /* although it doesn't seem documented, this code has long
12237 * no digits following the '.' is treated like '.0'
12238 * the number may be preceded by any number of zeroes,
12239 * e.g. "%.0001f", which is the same as "%.1f"
12240 * so I've kept that behaviour. DAPM May 2017
12244 precis = inRANGE(*q, '1', '9') ? expect_number(&q) : 0;
12253 case 'I': /* Ix, I32x, and I64x */
12254 # ifdef USE_64_BIT_INT
12255 if (q[1] == '6' && q[2] == '4') {
12261 if (q[1] == '3' && q[2] == '2') {
12265 # ifdef USE_64_BIT_INT
12271 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12272 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12275 # ifdef USE_QUADMATH
12288 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12289 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12290 if (*q == 'l') { /* lld, llf */
12299 if (*++q == 'h') { /* hhd, hhu */
12316 c = *q++; /* c now holds the conversion type */
12318 /* '%' doesn't have an arg, so skip arg processing */
12327 if (vectorize && !strchr("BbDdiOouUXx", c))
12330 /* get next arg (individual branches do their own va_arg()
12331 * handling for the args case) */
12334 efix = efix ? efix - 1 : svix++;
12335 argsv = efix < sv_count ? svargs[efix]
12336 : (arg_missing = TRUE, &PL_sv_no);
12346 eptr = va_arg(*args, char*);
12349 elen = my_strnlen(eptr, precis);
12351 elen = strlen(eptr);
12353 eptr = (char *)nullstr;
12354 elen = sizeof nullstr - 1;
12358 eptr = SvPV_const(argsv, elen);
12359 if (DO_UTF8(argsv)) {
12360 STRLEN old_precis = precis;
12361 if (has_precis && precis < elen) {
12362 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12363 STRLEN p = precis > ulen ? ulen : precis;
12364 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12365 /* sticks at end */
12367 if (width) { /* fudge width (can't fudge elen) */
12368 if (has_precis && precis < elen)
12369 width += precis - old_precis;
12372 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12379 if (has_precis && precis < elen)
12391 * "%...p" is normally treated like "%...x", except that the
12392 * number to print is the SV's address (or a pointer address
12393 * for C-ish sprintf).
12395 * However, the C-ish sprintf variant allows a few special
12396 * extensions. These are currently:
12398 * %-p (SVf) Like %s, but gets the string from an SV*
12399 * arg rather than a char* arg.
12400 * (This was previously %_).
12402 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12404 * %2p (HEKf) Like %s, but using the key string in a HEK
12406 * %3p (HEKf256) Ditto but like %.256s
12408 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12409 * (cBOOL(utf8), len, string_buf).
12410 * It's handled by the "case 'd'" branch
12411 * rather than here.
12413 * %<num>p where num is 1 or > 4: reserved for future
12414 * extensions. Warns, but then is treated as a
12415 * general %p (print hex address) format.
12423 /* not %*p or %*1$p - any width was explicit */
12427 if (left) { /* %-p (SVf), %-NNNp */
12432 argsv = MUTABLE_SV(va_arg(*args, void*));
12433 eptr = SvPV_const(argsv, elen);
12434 if (DO_UTF8(argsv))
12439 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12440 HEK * const hek = va_arg(*args, HEK *);
12441 eptr = HEK_KEY(hek);
12442 elen = HEK_LEN(hek);
12453 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12454 "internal %%<num>p might conflict with future printf extensions");
12458 /* treat as normal %...p */
12460 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12465 /* Ignore any size specifiers, since they're not documented as
12466 * being allowed for %c (ideally we should warn on e.g. '%hc').
12467 * Setting a default intsize, along with a positive
12468 * (which signals unsigned) base, causes, for C-ish use, the
12469 * va_arg to be interpreted as as unsigned int, when it's
12470 * actually signed, which will convert -ve values to high +ve
12471 * values. Note that unlike the libc %c, values > 255 will
12472 * convert to high unicode points rather than being truncated
12473 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12474 * will again convert -ve args to high -ve values.
12477 base = 1; /* special value that indicates we're doing a 'c' */
12478 goto get_int_arg_val;
12487 goto get_int_arg_val;
12490 /* probably just a plain %d, but it might be the start of the
12491 * special UTF8f format, which usually looks something like
12492 * "%d%lu%4p" (the lu may vary by platform)
12494 assert((UTF8f)[0] == 'd');
12495 assert((UTF8f)[1] == '%');
12497 if ( args /* UTF8f only valid for C-ish sprintf */
12498 && q == fmtstart + 1 /* plain %d, not %....d */
12499 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12501 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12503 /* The argument has already gone through cBOOL, so the cast
12505 is_utf8 = (bool)va_arg(*args, int);
12506 elen = va_arg(*args, UV);
12507 /* if utf8 length is larger than 0x7ffff..., then it might
12508 * have been a signed value that wrapped */
12509 if (elen > ((~(STRLEN)0) >> 1)) {
12510 assert(0); /* in DEBUGGING build we want to crash */
12511 elen = 0; /* otherwise we want to treat this as an empty string */
12513 eptr = va_arg(*args, char *);
12514 q += sizeof(UTF8f) - 2;
12521 goto get_int_arg_val;
12532 goto get_int_arg_val;
12537 goto get_int_arg_val;
12548 goto get_int_arg_val;
12563 esignbuf[esignlen++] = plus;
12566 /* initialise the vector string to iterate over */
12568 vecsv = args ? va_arg(*args, SV*) : argsv;
12570 /* if this is a version object, we need to convert
12571 * back into v-string notation and then let the
12572 * vectorize happen normally
12574 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12575 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12576 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12577 "vector argument not supported with alpha versions");
12581 vecstr = (U8*)SvPV_const(vecsv,veclen);
12582 vecsv = sv_newmortal();
12583 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12587 vecstr = (U8*)SvPV_const(vecsv, veclen);
12588 vec_utf8 = DO_UTF8(vecsv);
12590 /* This is the re-entry point for when we're iterating
12591 * over the individual characters of a vector arg */
12594 goto done_valid_conversion;
12596 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12606 /* test arg for inf/nan. This can trigger an unwanted
12607 * 'str' overload, so manually force 'num' overload first
12611 if (UNLIKELY(SvAMAGIC(argsv)))
12612 argsv = sv_2num(argsv);
12613 if (UNLIKELY(isinfnansv(argsv)))
12614 goto handle_infnan_argsv;
12618 /* signed int type */
12623 case 'c': iv = (char)va_arg(*args, int); break;
12624 case 'h': iv = (short)va_arg(*args, int); break;
12625 case 'l': iv = va_arg(*args, long); break;
12626 case 'V': iv = va_arg(*args, IV); break;
12627 case 'z': iv = va_arg(*args, SSize_t); break;
12628 #ifdef HAS_PTRDIFF_T
12629 case 't': iv = va_arg(*args, ptrdiff_t); break;
12631 default: iv = va_arg(*args, int); break;
12632 case 'j': iv = (IV) va_arg(*args, PERL_INTMAX_T); break;
12635 iv = va_arg(*args, Quad_t); break;
12642 /* assign to tiv then cast to iv to work around
12643 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12644 IV tiv = SvIV_nomg(argsv);
12646 case 'c': iv = (char)tiv; break;
12647 case 'h': iv = (short)tiv; break;
12648 case 'l': iv = (long)tiv; break;
12650 default: iv = tiv; break;
12653 iv = (Quad_t)tiv; break;
12660 /* now convert iv to uv */
12664 esignbuf[esignlen++] = plus;
12667 /* Using 0- here to silence bogus warning from MS VC */
12668 uv = (UV) (0 - (UV) iv);
12669 esignbuf[esignlen++] = '-';
12673 /* unsigned int type */
12676 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12678 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12680 case 'l': uv = va_arg(*args, unsigned long); break;
12681 case 'V': uv = va_arg(*args, UV); break;
12682 case 'z': uv = va_arg(*args, Size_t); break;
12683 #ifdef HAS_PTRDIFF_T
12684 /* will sign extend, but there is no
12685 * uptrdiff_t, so oh well */
12686 case 't': uv = va_arg(*args, ptrdiff_t); break;
12688 case 'j': uv = (UV) va_arg(*args, PERL_UINTMAX_T); break;
12689 default: uv = va_arg(*args, unsigned); break;
12692 uv = va_arg(*args, Uquad_t); break;
12699 /* assign to tiv then cast to iv to work around
12700 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12701 UV tuv = SvUV_nomg(argsv);
12703 case 'c': uv = (unsigned char)tuv; break;
12704 case 'h': uv = (unsigned short)tuv; break;
12705 case 'l': uv = (unsigned long)tuv; break;
12707 default: uv = tuv; break;
12710 uv = (Uquad_t)tuv; break;
12721 char *ptr = ebuf + sizeof ebuf;
12728 const char * const p =
12729 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12734 } while (uv >>= 4);
12735 if (alt && *ptr != '0') {
12736 esignbuf[esignlen++] = '0';
12737 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12744 *--ptr = '0' + dig;
12745 } while (uv >>= 3);
12746 if (alt && *ptr != '0')
12752 *--ptr = '0' + dig;
12753 } while (uv >>= 1);
12754 if (alt && *ptr != '0') {
12755 esignbuf[esignlen++] = '0';
12756 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12761 /* special-case: base 1 indicates a 'c' format:
12762 * we use the common code for extracting a uv,
12763 * but handle that value differently here than
12764 * all the other int types */
12766 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12769 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12771 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12776 ebuf[0] = (char)uv;
12781 default: /* it had better be ten or less */
12784 *--ptr = '0' + dig;
12785 } while (uv /= base);
12788 elen = (ebuf + sizeof ebuf) - ptr;
12792 zeros = precis - elen;
12793 else if (precis == 0 && elen == 1 && *eptr == '0'
12794 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12797 /* a precision nullifies the 0 flag. */
12803 /* FLOATING POINT */
12806 c = 'f'; /* maybe %F isn't supported here */
12808 case 'e': case 'E':
12810 case 'g': case 'G':
12811 case 'a': case 'A':
12814 STRLEN float_need; /* what PL_efloatsize needs to become */
12815 bool hexfp; /* hexadecimal floating point? */
12817 vcatpvfn_long_double_t fv;
12820 /* This is evil, but floating point is even more evil */
12822 /* for SV-style calling, we can only get NV
12823 for C-style calling, we assume %f is double;
12824 for simplicity we allow any of %Lf, %llf, %qf for long double
12828 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12832 /* [perl #20339] - we should accept and ignore %lf rather than die */
12836 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12837 intsize = args ? 0 : 'q';
12841 #if defined(HAS_LONG_DOUBLE)
12854 /* Now we need (long double) if intsize == 'q', else (double). */
12856 /* Note: do not pull NVs off the va_list with va_arg()
12857 * (pull doubles instead) because if you have a build
12858 * with long doubles, you would always be pulling long
12859 * doubles, which would badly break anyone using only
12860 * doubles (i.e. the majority of builds). In other
12861 * words, you cannot mix doubles and long doubles.
12862 * The only case where you can pull off long doubles
12863 * is when the format specifier explicitly asks so with
12865 #ifdef USE_QUADMATH
12866 fv = intsize == 'q' ?
12867 va_arg(*args, NV) : va_arg(*args, double);
12869 #elif LONG_DOUBLESIZE > DOUBLESIZE
12870 if (intsize == 'q') {
12871 fv = va_arg(*args, long double);
12874 nv = va_arg(*args, double);
12875 VCATPVFN_NV_TO_FV(nv, fv);
12878 nv = va_arg(*args, double);
12885 /* we jump here if an int-ish format encountered an
12886 * infinite/Nan argsv. After setting nv/fv, it falls
12887 * into the isinfnan block which follows */
12888 handle_infnan_argsv:
12889 nv = SvNV_nomg(argsv);
12890 VCATPVFN_NV_TO_FV(nv, fv);
12893 if (Perl_isinfnan(nv)) {
12895 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12896 SvNV_nomg(argsv), (int)c);
12898 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12907 /* special-case "%.0f" */
12911 && !(width || left || plus || alt)
12914 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12918 /* Determine the buffer size needed for the various
12919 * floating-point formats.
12921 * The basic possibilities are:
12924 * %f 1111111.123456789
12925 * %e 1.111111123e+06
12926 * %a 0x1.0f4471f9bp+20
12928 * %g 1.11111112e+15
12930 * where P is the value of the precision in the format, or 6
12931 * if not specified. Note the two possible output formats of
12932 * %g; in both cases the number of significant digits is <=
12935 * For most of the format types the maximum buffer size needed
12936 * is precision, plus: any leading 1 or 0x1, the radix
12937 * point, and an exponent. The difficult one is %f: for a
12938 * large positive exponent it can have many leading digits,
12939 * which needs to be calculated specially. Also %a is slightly
12940 * different in that in the absence of a specified precision,
12941 * it uses as many digits as necessary to distinguish
12942 * different values.
12944 * First, here are the constant bits. For ease of calculation
12945 * we over-estimate the needed buffer size, for example by
12946 * assuming all formats have an exponent and a leading 0x1.
12948 * Also for production use, add a little extra overhead for
12949 * safety's sake. Under debugging don't, as it means we're
12950 * more likely to quickly spot issues during development.
12953 float_need = 1 /* possible unary minus */
12954 + 4 /* "0x1" plus very unlikely carry */
12955 + 1 /* default radix point '.' */
12956 + 2 /* "e-", "p+" etc */
12957 + 6 /* exponent: up to 16383 (quad fp) */
12959 + 20 /* safety net */
12964 /* determine the radix point len, e.g. length(".") in "1.2" */
12965 #ifdef USE_LOCALE_NUMERIC
12966 /* note that we may either explicitly use PL_numeric_radix_sv
12967 * below, or implicitly, via an snprintf() variant.
12968 * Note also things like ps_AF.utf8 which has
12969 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
12970 if (!lc_numeric_set) {
12971 /* only set once and reuse in-locale value on subsequent
12973 * XXX what happens if we die in an eval?
12975 STORE_LC_NUMERIC_SET_TO_NEEDED();
12976 lc_numeric_set = TRUE;
12979 if (IN_LC(LC_NUMERIC)) {
12980 /* this can't wrap unless PL_numeric_radix_sv is a string
12981 * consuming virtually all the 32-bit or 64-bit address
12984 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
12986 /* floating-point formats only get utf8 if the radix point
12987 * is utf8. All other characters in the string are < 128
12988 * and so can be safely appended to both a non-utf8 and utf8
12990 * Note that this will convert the output to utf8 even if
12991 * the radix point didn't get output.
12993 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
12994 sv_utf8_upgrade(sv);
13002 if (isALPHA_FOLD_EQ(c, 'f')) {
13003 /* Determine how many digits before the radix point
13004 * might be emitted. frexp() (or frexpl) has some
13005 * unspecified behaviour for nan/inf/-inf, so lucky we've
13006 * already handled them above */
13008 int i = PERL_INT_MIN;
13009 (void)Perl_frexp((NV)fv, &i);
13010 if (i == PERL_INT_MIN)
13011 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13014 digits = BIT_DIGITS(i);
13015 /* this can't overflow. 'digits' will only be a few
13016 * thousand even for the largest floating-point types.
13017 * And up until now float_need is just some small
13018 * constants plus radix len, which can't be in
13019 * overflow territory unless the radix SV is consuming
13020 * over 1/2 the address space */
13021 assert(float_need < ((STRLEN)~0) - digits);
13022 float_need += digits;
13025 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13028 /* %a in the absence of precision may print as many
13029 * digits as needed to represent the entire mantissa
13031 * This estimate seriously overshoots in most cases,
13032 * but better the undershooting. Firstly, all bytes
13033 * of the NV are not mantissa, some of them are
13034 * exponent. Secondly, for the reasonably common
13035 * long doubles case, the "80-bit extended", two
13036 * or six bytes of the NV are unused. Also, we'll
13037 * still pick up an extra +6 from the default
13038 * precision calculation below. */
13040 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13041 /* For the "double double", we need more.
13042 * Since each double has their own exponent, the
13043 * doubles may float (haha) rather far from each
13044 * other, and the number of required bits is much
13045 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13046 * See the definition of DOUBLEDOUBLE_MAXBITS.
13048 * Need 2 hexdigits for each byte. */
13049 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13051 NVSIZE * 2; /* 2 hexdigits for each byte */
13053 /* see "this can't overflow" comment above */
13054 assert(float_need < ((STRLEN)~0) - digits);
13055 float_need += digits;
13058 /* special-case "%.<number>g" if it will fit in ebuf */
13060 && precis /* See earlier comment about buggy Gconvert
13061 when digits, aka precis, is 0 */
13063 /* check, in manner not involving wrapping, that it will
13065 && float_need < sizeof(ebuf)
13066 && sizeof(ebuf) - float_need > precis
13067 && !(width || left || plus || alt)
13071 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13072 elen = strlen(ebuf);
13079 STRLEN pr = has_precis ? precis : 6; /* known default */
13080 /* this probably can't wrap, since precis is limited
13081 * to 1/4 address space size, but better safe than sorry
13083 if (float_need >= ((STRLEN)~0) - pr)
13084 croak_memory_wrap();
13088 if (float_need < width)
13089 float_need = width;
13091 if (float_need > INT_MAX) {
13092 /* snprintf() returns an int, and we use that return value,
13093 so die horribly if the expected size is too large for int
13095 Perl_croak(aTHX_ "Numeric format result too large");
13098 if (PL_efloatsize <= float_need) {
13099 /* PL_efloatbuf should be at least 1 greater than
13100 * float_need to allow a trailing \0 to be returned by
13101 * snprintf(). If we need to grow, overgrow for the
13102 * benefit of future generations */
13103 const STRLEN extra = 0x20;
13104 if (float_need >= ((STRLEN)~0) - extra)
13105 croak_memory_wrap();
13106 float_need += extra;
13107 Safefree(PL_efloatbuf);
13108 PL_efloatsize = float_need;
13109 Newx(PL_efloatbuf, PL_efloatsize, char);
13110 PL_efloatbuf[0] = '\0';
13113 if (UNLIKELY(hexfp)) {
13114 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13115 nv, fv, has_precis, precis, width,
13116 alt, plus, left, fill);
13119 char *ptr = ebuf + sizeof ebuf;
13122 #if defined(USE_QUADMATH)
13123 if (intsize == 'q') {
13127 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13128 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13129 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13130 * not USE_LONG_DOUBLE and NVff. In other words,
13131 * this needs to work without USE_LONG_DOUBLE. */
13132 if (intsize == 'q') {
13133 /* Copy the one or more characters in a long double
13134 * format before the 'base' ([efgEFG]) character to
13135 * the format string. */
13136 static char const ldblf[] = PERL_PRIfldbl;
13137 char const *p = ldblf + sizeof(ldblf) - 3;
13138 while (p >= ldblf) { *--ptr = *p--; }
13143 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13148 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13160 /* No taint. Otherwise we are in the strange situation
13161 * where printf() taints but print($float) doesn't.
13164 /* hopefully the above makes ptr a very constrained format
13165 * that is safe to use, even though it's not literal */
13166 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13167 #ifdef USE_QUADMATH
13169 const char* qfmt = quadmath_format_single(ptr);
13171 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13172 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13174 if ((IV)elen == -1) {
13177 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13182 #elif defined(HAS_LONG_DOUBLE)
13183 elen = ((intsize == 'q')
13184 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13185 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13187 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13189 GCC_DIAG_RESTORE_STMT;
13192 eptr = PL_efloatbuf;
13196 /* Since floating-point formats do their own formatting and
13197 * padding, we skip the main block of code at the end of this
13198 * loop which handles appending eptr to sv, and do our own
13199 * stripped-down version */
13204 assert(elen >= width);
13206 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13208 goto done_valid_conversion;
13216 /* XXX ideally we should warn if any flags etc have been
13217 * set, e.g. "%-4.5n" */
13218 /* XXX if sv was originally non-utf8 with a char in the
13219 * range 0x80-0xff, then if it got upgraded, we should
13220 * calculate char len rather than byte len here */
13221 len = SvCUR(sv) - origlen;
13223 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13226 case 'c': *(va_arg(*args, char*)) = i; break;
13227 case 'h': *(va_arg(*args, short*)) = i; break;
13228 default: *(va_arg(*args, int*)) = i; break;
13229 case 'l': *(va_arg(*args, long*)) = i; break;
13230 case 'V': *(va_arg(*args, IV*)) = i; break;
13231 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13232 #ifdef HAS_PTRDIFF_T
13233 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13235 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13238 *(va_arg(*args, Quad_t*)) = i; break;
13246 Perl_croak_nocontext(
13247 "Missing argument for %%n in %s",
13248 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13249 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13251 goto done_valid_conversion;
13259 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13260 && ckWARN(WARN_PRINTF))
13262 SV * const msg = sv_newmortal();
13263 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13264 (PL_op->op_type == OP_PRTF) ? "" : "s");
13265 if (fmtstart < patend) {
13266 const char * const fmtend = q < patend ? q : patend;
13268 sv_catpvs(msg, "\"%");
13269 for (f = fmtstart; f < fmtend; f++) {
13271 sv_catpvn_nomg(msg, f, 1);
13273 Perl_sv_catpvf(aTHX_ msg,
13274 "\\%03" UVof, (UV)*f & 0xFF);
13277 sv_catpvs(msg, "\"");
13279 sv_catpvs(msg, "end of string");
13281 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13284 /* mangled format: output the '%', then continue from the
13285 * character following that */
13286 sv_catpvn_nomg(sv, fmtstart-1, 1);
13289 /* Any "redundant arg" warning from now onwards will probably
13290 * just be misleading, so don't bother. */
13291 no_redundant_warning = TRUE;
13292 continue; /* not "break" */
13295 if (is_utf8 != has_utf8) {
13298 sv_utf8_upgrade(sv);
13301 const STRLEN old_elen = elen;
13302 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13303 sv_utf8_upgrade(nsv);
13304 eptr = SvPVX_const(nsv);
13307 if (width) { /* fudge width (can't fudge elen) */
13308 width += elen - old_elen;
13315 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13318 STRLEN need, have, gap;
13322 /* signed value that's wrapped? */
13323 assert(elen <= ((~(STRLEN)0) >> 1));
13325 /* if zeros is non-zero, then it represents filler between
13326 * elen and precis. So adding elen and zeros together will
13327 * always be <= precis, and the addition can never wrap */
13328 assert(!zeros || (precis > elen && precis - elen == zeros));
13329 have = elen + zeros;
13331 if (have >= (((STRLEN)~0) - esignlen))
13332 croak_memory_wrap();
13335 need = (have > width ? have : width);
13338 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13339 croak_memory_wrap();
13340 need += (SvCUR(sv) + 1);
13347 for (i = 0; i < esignlen; i++)
13348 *s++ = esignbuf[i];
13349 for (i = zeros; i; i--)
13351 Copy(eptr, s, elen, char);
13353 for (i = gap; i; i--)
13358 for (i = 0; i < esignlen; i++)
13359 *s++ = esignbuf[i];
13364 for (i = gap; i; i--)
13366 for (i = 0; i < esignlen; i++)
13367 *s++ = esignbuf[i];
13370 for (i = zeros; i; i--)
13372 Copy(eptr, s, elen, char);
13377 SvCUR_set(sv, s - SvPVX_const(sv));
13385 if (vectorize && veclen) {
13386 /* we append the vector separator separately since %v isn't
13387 * very common: don't slow down the general case by adding
13388 * dotstrlen to need etc */
13389 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13391 goto vector; /* do next iteration */
13394 done_valid_conversion:
13397 S_warn_vcatpvfn_missing_argument(aTHX);
13400 /* Now that we've consumed all our printf format arguments (svix)
13401 * do we have things left on the stack that we didn't use?
13403 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13404 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13405 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13410 #ifdef USE_LOCALE_NUMERIC
13412 if (lc_numeric_set) {
13413 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to
13414 save/restore each iteration. */
13421 /* =========================================================================
13423 =head1 Cloning an interpreter
13427 All the macros and functions in this section are for the private use of
13428 the main function, perl_clone().
13430 The foo_dup() functions make an exact copy of an existing foo thingy.
13431 During the course of a cloning, a hash table is used to map old addresses
13432 to new addresses. The table is created and manipulated with the
13433 ptr_table_* functions.
13435 * =========================================================================*/
13438 #if defined(USE_ITHREADS)
13440 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13441 #ifndef GpREFCNT_inc
13442 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13446 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13447 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13448 If this changes, please unmerge ss_dup.
13449 Likewise, sv_dup_inc_multiple() relies on this fact. */
13450 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13451 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13452 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13453 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13454 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13455 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13456 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13457 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13458 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13459 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13460 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13461 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13462 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13464 /* clone a parser */
13467 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13471 PERL_ARGS_ASSERT_PARSER_DUP;
13476 /* look for it in the table first */
13477 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13481 /* create anew and remember what it is */
13482 Newxz(parser, 1, yy_parser);
13483 ptr_table_store(PL_ptr_table, proto, parser);
13485 /* XXX eventually, just Copy() most of the parser struct ? */
13487 parser->lex_brackets = proto->lex_brackets;
13488 parser->lex_casemods = proto->lex_casemods;
13489 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13490 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13491 parser->lex_casestack = savepvn(proto->lex_casestack,
13492 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13493 parser->lex_defer = proto->lex_defer;
13494 parser->lex_dojoin = proto->lex_dojoin;
13495 parser->lex_formbrack = proto->lex_formbrack;
13496 parser->lex_inpat = proto->lex_inpat;
13497 parser->lex_inwhat = proto->lex_inwhat;
13498 parser->lex_op = proto->lex_op;
13499 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13500 parser->lex_starts = proto->lex_starts;
13501 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13502 parser->multi_close = proto->multi_close;
13503 parser->multi_open = proto->multi_open;
13504 parser->multi_start = proto->multi_start;
13505 parser->multi_end = proto->multi_end;
13506 parser->preambled = proto->preambled;
13507 parser->lex_super_state = proto->lex_super_state;
13508 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13509 parser->lex_sub_op = proto->lex_sub_op;
13510 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13511 parser->linestr = sv_dup_inc(proto->linestr, param);
13512 parser->expect = proto->expect;
13513 parser->copline = proto->copline;
13514 parser->last_lop_op = proto->last_lop_op;
13515 parser->lex_state = proto->lex_state;
13516 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13517 /* rsfp_filters entries have fake IoDIRP() */
13518 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13519 parser->in_my = proto->in_my;
13520 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13521 parser->error_count = proto->error_count;
13522 parser->sig_elems = proto->sig_elems;
13523 parser->sig_optelems= proto->sig_optelems;
13524 parser->sig_slurpy = proto->sig_slurpy;
13525 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13528 char * const ols = SvPVX(proto->linestr);
13529 char * const ls = SvPVX(parser->linestr);
13531 parser->bufptr = ls + (proto->bufptr >= ols ?
13532 proto->bufptr - ols : 0);
13533 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13534 proto->oldbufptr - ols : 0);
13535 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13536 proto->oldoldbufptr - ols : 0);
13537 parser->linestart = ls + (proto->linestart >= ols ?
13538 proto->linestart - ols : 0);
13539 parser->last_uni = ls + (proto->last_uni >= ols ?
13540 proto->last_uni - ols : 0);
13541 parser->last_lop = ls + (proto->last_lop >= ols ?
13542 proto->last_lop - ols : 0);
13544 parser->bufend = ls + SvCUR(parser->linestr);
13547 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13550 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13551 Copy(proto->nexttype, parser->nexttype, 5, I32);
13552 parser->nexttoke = proto->nexttoke;
13554 /* XXX should clone saved_curcop here, but we aren't passed
13555 * proto_perl; so do it in perl_clone_using instead */
13561 /* duplicate a file handle */
13564 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13568 PERL_ARGS_ASSERT_FP_DUP;
13569 PERL_UNUSED_ARG(type);
13572 return (PerlIO*)NULL;
13574 /* look for it in the table first */
13575 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13579 /* create anew and remember what it is */
13580 #ifdef __amigaos4__
13581 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13583 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13585 ptr_table_store(PL_ptr_table, fp, ret);
13589 /* duplicate a directory handle */
13592 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13596 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13598 const Direntry_t *dirent;
13599 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13605 PERL_UNUSED_CONTEXT;
13606 PERL_ARGS_ASSERT_DIRP_DUP;
13611 /* look for it in the table first */
13612 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13616 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13618 PERL_UNUSED_ARG(param);
13622 /* open the current directory (so we can switch back) */
13623 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13625 /* chdir to our dir handle and open the present working directory */
13626 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13627 PerlDir_close(pwd);
13628 return (DIR *)NULL;
13630 /* Now we should have two dir handles pointing to the same dir. */
13632 /* Be nice to the calling code and chdir back to where we were. */
13633 /* XXX If this fails, then what? */
13634 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13636 /* We have no need of the pwd handle any more. */
13637 PerlDir_close(pwd);
13640 # define d_namlen(d) (d)->d_namlen
13642 # define d_namlen(d) strlen((d)->d_name)
13644 /* Iterate once through dp, to get the file name at the current posi-
13645 tion. Then step back. */
13646 pos = PerlDir_tell(dp);
13647 if ((dirent = PerlDir_read(dp))) {
13648 len = d_namlen(dirent);
13649 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13650 /* If the len is somehow magically longer than the
13651 * maximum length of the directory entry, even though
13652 * we could fit it in a buffer, we could not copy it
13653 * from the dirent. Bail out. */
13654 PerlDir_close(ret);
13657 if (len <= sizeof smallbuf) name = smallbuf;
13658 else Newx(name, len, char);
13659 Move(dirent->d_name, name, len, char);
13661 PerlDir_seek(dp, pos);
13663 /* Iterate through the new dir handle, till we find a file with the
13665 if (!dirent) /* just before the end */
13667 pos = PerlDir_tell(ret);
13668 if (PerlDir_read(ret)) continue; /* not there yet */
13669 PerlDir_seek(ret, pos); /* step back */
13673 const long pos0 = PerlDir_tell(ret);
13675 pos = PerlDir_tell(ret);
13676 if ((dirent = PerlDir_read(ret))) {
13677 if (len == (STRLEN)d_namlen(dirent)
13678 && memEQ(name, dirent->d_name, len)) {
13680 PerlDir_seek(ret, pos); /* step back */
13683 /* else we are not there yet; keep iterating */
13685 else { /* This is not meant to happen. The best we can do is
13686 reset the iterator to the beginning. */
13687 PerlDir_seek(ret, pos0);
13694 if (name && name != smallbuf)
13699 ret = win32_dirp_dup(dp, param);
13702 /* pop it in the pointer table */
13704 ptr_table_store(PL_ptr_table, dp, ret);
13709 /* duplicate a typeglob */
13712 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13716 PERL_ARGS_ASSERT_GP_DUP;
13720 /* look for it in the table first */
13721 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13725 /* create anew and remember what it is */
13727 ptr_table_store(PL_ptr_table, gp, ret);
13730 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13731 on Newxz() to do this for us. */
13732 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13733 ret->gp_io = io_dup_inc(gp->gp_io, param);
13734 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13735 ret->gp_av = av_dup_inc(gp->gp_av, param);
13736 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13737 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13738 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13739 ret->gp_cvgen = gp->gp_cvgen;
13740 ret->gp_line = gp->gp_line;
13741 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13745 /* duplicate a chain of magic */
13748 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13750 MAGIC *mgret = NULL;
13751 MAGIC **mgprev_p = &mgret;
13753 PERL_ARGS_ASSERT_MG_DUP;
13755 for (; mg; mg = mg->mg_moremagic) {
13758 if ((param->flags & CLONEf_JOIN_IN)
13759 && mg->mg_type == PERL_MAGIC_backref)
13760 /* when joining, we let the individual SVs add themselves to
13761 * backref as needed. */
13764 Newx(nmg, 1, MAGIC);
13766 mgprev_p = &(nmg->mg_moremagic);
13768 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13769 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13770 from the original commit adding Perl_mg_dup() - revision 4538.
13771 Similarly there is the annotation "XXX random ptr?" next to the
13772 assignment to nmg->mg_ptr. */
13775 /* FIXME for plugins
13776 if (nmg->mg_type == PERL_MAGIC_qr) {
13777 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13781 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13782 ? nmg->mg_type == PERL_MAGIC_backref
13783 /* The backref AV has its reference
13784 * count deliberately bumped by 1 */
13785 ? SvREFCNT_inc(av_dup_inc((const AV *)
13786 nmg->mg_obj, param))
13787 : sv_dup_inc(nmg->mg_obj, param)
13788 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13789 nmg->mg_type == PERL_MAGIC_regdata)
13791 : sv_dup(nmg->mg_obj, param);
13793 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13794 if (nmg->mg_len > 0) {
13795 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13796 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13797 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13799 AMT * const namtp = (AMT*)nmg->mg_ptr;
13800 sv_dup_inc_multiple((SV**)(namtp->table),
13801 (SV**)(namtp->table), NofAMmeth, param);
13804 else if (nmg->mg_len == HEf_SVKEY)
13805 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13807 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13808 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13814 #endif /* USE_ITHREADS */
13816 struct ptr_tbl_arena {
13817 struct ptr_tbl_arena *next;
13818 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13821 /* create a new pointer-mapping table */
13824 Perl_ptr_table_new(pTHX)
13827 PERL_UNUSED_CONTEXT;
13829 Newx(tbl, 1, PTR_TBL_t);
13830 tbl->tbl_max = 511;
13831 tbl->tbl_items = 0;
13832 tbl->tbl_arena = NULL;
13833 tbl->tbl_arena_next = NULL;
13834 tbl->tbl_arena_end = NULL;
13835 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13839 #define PTR_TABLE_HASH(ptr) \
13840 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13842 /* map an existing pointer using a table */
13844 STATIC PTR_TBL_ENT_t *
13845 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13847 PTR_TBL_ENT_t *tblent;
13848 const UV hash = PTR_TABLE_HASH(sv);
13850 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13852 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13853 for (; tblent; tblent = tblent->next) {
13854 if (tblent->oldval == sv)
13861 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13863 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13865 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13866 PERL_UNUSED_CONTEXT;
13868 return tblent ? tblent->newval : NULL;
13871 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13872 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13873 * the core's typical use of ptr_tables in thread cloning. */
13876 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13878 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13880 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13881 PERL_UNUSED_CONTEXT;
13884 tblent->newval = newsv;
13886 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13888 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13889 struct ptr_tbl_arena *new_arena;
13891 Newx(new_arena, 1, struct ptr_tbl_arena);
13892 new_arena->next = tbl->tbl_arena;
13893 tbl->tbl_arena = new_arena;
13894 tbl->tbl_arena_next = new_arena->array;
13895 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13898 tblent = tbl->tbl_arena_next++;
13900 tblent->oldval = oldsv;
13901 tblent->newval = newsv;
13902 tblent->next = tbl->tbl_ary[entry];
13903 tbl->tbl_ary[entry] = tblent;
13905 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13906 ptr_table_split(tbl);
13910 /* double the hash bucket size of an existing ptr table */
13913 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13915 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13916 const UV oldsize = tbl->tbl_max + 1;
13917 UV newsize = oldsize * 2;
13920 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13921 PERL_UNUSED_CONTEXT;
13923 Renew(ary, newsize, PTR_TBL_ENT_t*);
13924 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13925 tbl->tbl_max = --newsize;
13926 tbl->tbl_ary = ary;
13927 for (i=0; i < oldsize; i++, ary++) {
13928 PTR_TBL_ENT_t **entp = ary;
13929 PTR_TBL_ENT_t *ent = *ary;
13930 PTR_TBL_ENT_t **curentp;
13933 curentp = ary + oldsize;
13935 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13937 ent->next = *curentp;
13947 /* remove all the entries from a ptr table */
13948 /* Deprecated - will be removed post 5.14 */
13951 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13953 PERL_UNUSED_CONTEXT;
13954 if (tbl && tbl->tbl_items) {
13955 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13957 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13960 struct ptr_tbl_arena *next = arena->next;
13966 tbl->tbl_items = 0;
13967 tbl->tbl_arena = NULL;
13968 tbl->tbl_arena_next = NULL;
13969 tbl->tbl_arena_end = NULL;
13973 /* clear and free a ptr table */
13976 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13978 struct ptr_tbl_arena *arena;
13980 PERL_UNUSED_CONTEXT;
13986 arena = tbl->tbl_arena;
13989 struct ptr_tbl_arena *next = arena->next;
13995 Safefree(tbl->tbl_ary);
13999 #if defined(USE_ITHREADS)
14002 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14004 PERL_ARGS_ASSERT_RVPV_DUP;
14006 assert(!isREGEXP(sstr));
14008 if (SvWEAKREF(sstr)) {
14009 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14010 if (param->flags & CLONEf_JOIN_IN) {
14011 /* if joining, we add any back references individually rather
14012 * than copying the whole backref array */
14013 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14017 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14019 else if (SvPVX_const(sstr)) {
14020 /* Has something there */
14022 /* Normal PV - clone whole allocated space */
14023 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14024 /* sstr may not be that normal, but actually copy on write.
14025 But we are a true, independent SV, so: */
14029 /* Special case - not normally malloced for some reason */
14030 if (isGV_with_GP(sstr)) {
14031 /* Don't need to do anything here. */
14033 else if ((SvIsCOW(sstr))) {
14034 /* A "shared" PV - clone it as "shared" PV */
14036 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14040 /* Some other special case - random pointer */
14041 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14046 /* Copy the NULL */
14047 SvPV_set(dstr, NULL);
14051 /* duplicate a list of SVs. source and dest may point to the same memory. */
14053 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14054 SSize_t items, CLONE_PARAMS *const param)
14056 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14058 while (items-- > 0) {
14059 *dest++ = sv_dup_inc(*source++, param);
14065 /* duplicate an SV of any type (including AV, HV etc) */
14068 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14073 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14075 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14076 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14081 /* look for it in the table first */
14082 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14086 if(param->flags & CLONEf_JOIN_IN) {
14087 /** We are joining here so we don't want do clone
14088 something that is bad **/
14089 if (SvTYPE(sstr) == SVt_PVHV) {
14090 const HEK * const hvname = HvNAME_HEK(sstr);
14092 /** don't clone stashes if they already exist **/
14093 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14094 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14095 ptr_table_store(PL_ptr_table, sstr, dstr);
14099 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14100 HV *stash = GvSTASH(sstr);
14101 const HEK * hvname;
14102 if (stash && (hvname = HvNAME_HEK(stash))) {
14103 /** don't clone GVs if they already exist **/
14105 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14106 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14108 stash, GvNAME(sstr),
14114 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14115 ptr_table_store(PL_ptr_table, sstr, *svp);
14122 /* create anew and remember what it is */
14125 #ifdef DEBUG_LEAKING_SCALARS
14126 dstr->sv_debug_optype = sstr->sv_debug_optype;
14127 dstr->sv_debug_line = sstr->sv_debug_line;
14128 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14129 dstr->sv_debug_parent = (SV*)sstr;
14130 FREE_SV_DEBUG_FILE(dstr);
14131 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14134 ptr_table_store(PL_ptr_table, sstr, dstr);
14137 SvFLAGS(dstr) = SvFLAGS(sstr);
14138 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14139 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14142 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14143 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14144 (void*)PL_watch_pvx, SvPVX_const(sstr));
14147 /* don't clone objects whose class has asked us not to */
14149 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14155 switch (SvTYPE(sstr)) {
14157 SvANY(dstr) = NULL;
14160 SET_SVANY_FOR_BODYLESS_IV(dstr);
14162 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14164 SvIV_set(dstr, SvIVX(sstr));
14168 #if NVSIZE <= IVSIZE
14169 SET_SVANY_FOR_BODYLESS_NV(dstr);
14171 SvANY(dstr) = new_XNV();
14173 SvNV_set(dstr, SvNVX(sstr));
14177 /* These are all the types that need complex bodies allocating. */
14179 const svtype sv_type = SvTYPE(sstr);
14180 const struct body_details *const sv_type_details
14181 = bodies_by_type + sv_type;
14185 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14186 NOT_REACHED; /* NOTREACHED */
14202 assert(sv_type_details->body_size);
14203 if (sv_type_details->arena) {
14204 new_body_inline(new_body, sv_type);
14206 = (void*)((char*)new_body - sv_type_details->offset);
14208 new_body = new_NOARENA(sv_type_details);
14212 SvANY(dstr) = new_body;
14215 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14216 ((char*)SvANY(dstr)) + sv_type_details->offset,
14217 sv_type_details->copy, char);
14219 Copy(((char*)SvANY(sstr)),
14220 ((char*)SvANY(dstr)),
14221 sv_type_details->body_size + sv_type_details->offset, char);
14224 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14225 && !isGV_with_GP(dstr)
14227 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14228 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14230 /* The Copy above means that all the source (unduplicated) pointers
14231 are now in the destination. We can check the flags and the
14232 pointers in either, but it's possible that there's less cache
14233 missing by always going for the destination.
14234 FIXME - instrument and check that assumption */
14235 if (sv_type >= SVt_PVMG) {
14237 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14238 if (SvOBJECT(dstr) && SvSTASH(dstr))
14239 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14240 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14243 /* The cast silences a GCC warning about unhandled types. */
14244 switch ((int)sv_type) {
14255 /* FIXME for plugins */
14256 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14259 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14260 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14261 LvTARG(dstr) = dstr;
14262 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14263 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14265 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14266 if (isREGEXP(sstr)) goto duprex;
14269 /* non-GP case already handled above */
14270 if(isGV_with_GP(sstr)) {
14271 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14272 /* Don't call sv_add_backref here as it's going to be
14273 created as part of the magic cloning of the symbol
14274 table--unless this is during a join and the stash
14275 is not actually being cloned. */
14276 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14277 at the point of this comment. */
14278 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14279 if (param->flags & CLONEf_JOIN_IN)
14280 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14281 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14282 (void)GpREFCNT_inc(GvGP(dstr));
14286 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14287 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14288 /* I have no idea why fake dirp (rsfps)
14289 should be treated differently but otherwise
14290 we end up with leaks -- sky*/
14291 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14292 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14293 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14295 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14296 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14297 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14298 if (IoDIRP(dstr)) {
14299 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14302 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14304 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14306 if (IoOFP(dstr) == IoIFP(sstr))
14307 IoOFP(dstr) = IoIFP(dstr);
14309 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14310 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14311 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14312 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14315 /* avoid cloning an empty array */
14316 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14317 SV **dst_ary, **src_ary;
14318 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14320 src_ary = AvARRAY((const AV *)sstr);
14321 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14322 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14323 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14324 AvALLOC((const AV *)dstr) = dst_ary;
14325 if (AvREAL((const AV *)sstr)) {
14326 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14330 while (items-- > 0)
14331 *dst_ary++ = sv_dup(*src_ary++, param);
14333 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14334 while (items-- > 0) {
14339 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14340 AvALLOC((const AV *)dstr) = (SV**)NULL;
14341 AvMAX( (const AV *)dstr) = -1;
14342 AvFILLp((const AV *)dstr) = -1;
14346 if (HvARRAY((const HV *)sstr)) {
14348 const bool sharekeys = !!HvSHAREKEYS(sstr);
14349 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14350 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14352 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14353 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14355 HvARRAY(dstr) = (HE**)darray;
14356 while (i <= sxhv->xhv_max) {
14357 const HE * const source = HvARRAY(sstr)[i];
14358 HvARRAY(dstr)[i] = source
14359 ? he_dup(source, sharekeys, param) : 0;
14363 const struct xpvhv_aux * const saux = HvAUX(sstr);
14364 struct xpvhv_aux * const daux = HvAUX(dstr);
14365 /* This flag isn't copied. */
14368 if (saux->xhv_name_count) {
14369 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14371 = saux->xhv_name_count < 0
14372 ? -saux->xhv_name_count
14373 : saux->xhv_name_count;
14374 HEK **shekp = sname + count;
14376 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14377 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14378 while (shekp-- > sname) {
14380 *dhekp = hek_dup(*shekp, param);
14384 daux->xhv_name_u.xhvnameu_name
14385 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14388 daux->xhv_name_count = saux->xhv_name_count;
14390 daux->xhv_aux_flags = saux->xhv_aux_flags;
14391 #ifdef PERL_HASH_RANDOMIZE_KEYS
14392 daux->xhv_rand = saux->xhv_rand;
14393 daux->xhv_last_rand = saux->xhv_last_rand;
14395 daux->xhv_riter = saux->xhv_riter;
14396 daux->xhv_eiter = saux->xhv_eiter
14397 ? he_dup(saux->xhv_eiter,
14398 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14399 /* backref array needs refcnt=2; see sv_add_backref */
14400 daux->xhv_backreferences =
14401 (param->flags & CLONEf_JOIN_IN)
14402 /* when joining, we let the individual GVs and
14403 * CVs add themselves to backref as
14404 * needed. This avoids pulling in stuff
14405 * that isn't required, and simplifies the
14406 * case where stashes aren't cloned back
14407 * if they already exist in the parent
14410 : saux->xhv_backreferences
14411 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14412 ? MUTABLE_AV(SvREFCNT_inc(
14413 sv_dup_inc((const SV *)
14414 saux->xhv_backreferences, param)))
14415 : MUTABLE_AV(sv_dup((const SV *)
14416 saux->xhv_backreferences, param))
14419 daux->xhv_mro_meta = saux->xhv_mro_meta
14420 ? mro_meta_dup(saux->xhv_mro_meta, param)
14423 /* Record stashes for possible cloning in Perl_clone(). */
14425 av_push(param->stashes, dstr);
14429 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14432 if (!(param->flags & CLONEf_COPY_STACKS)) {
14437 /* NOTE: not refcounted */
14438 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14439 hv_dup(CvSTASH(dstr), param);
14440 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14441 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14442 if (!CvISXSUB(dstr)) {
14444 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14446 CvSLABBED_off(dstr);
14447 } else if (CvCONST(dstr)) {
14448 CvXSUBANY(dstr).any_ptr =
14449 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14451 assert(!CvSLABBED(dstr));
14452 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14454 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14455 hek_dup(CvNAME_HEK((CV *)sstr), param);
14456 /* don't dup if copying back - CvGV isn't refcounted, so the
14457 * duped GV may never be freed. A bit of a hack! DAPM */
14459 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14461 ? gv_dup_inc(CvGV(sstr), param)
14462 : (param->flags & CLONEf_JOIN_IN)
14464 : gv_dup(CvGV(sstr), param);
14466 if (!CvISXSUB(sstr)) {
14467 PADLIST * padlist = CvPADLIST(sstr);
14469 padlist = padlist_dup(padlist, param);
14470 CvPADLIST_set(dstr, padlist);
14472 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14473 PoisonPADLIST(dstr);
14476 CvWEAKOUTSIDE(sstr)
14477 ? cv_dup( CvOUTSIDE(dstr), param)
14478 : cv_dup_inc(CvOUTSIDE(dstr), param);
14488 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14490 PERL_ARGS_ASSERT_SV_DUP_INC;
14491 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14495 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14497 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14498 PERL_ARGS_ASSERT_SV_DUP;
14500 /* Track every SV that (at least initially) had a reference count of 0.
14501 We need to do this by holding an actual reference to it in this array.
14502 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14503 (akin to the stashes hash, and the perl stack), we come unstuck if
14504 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14505 thread) is manipulated in a CLONE method, because CLONE runs before the
14506 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14507 (and fix things up by giving each a reference via the temps stack).
14508 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14509 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14510 before the walk of unreferenced happens and a reference to that is SV
14511 added to the temps stack. At which point we have the same SV considered
14512 to be in use, and free to be re-used. Not good.
14514 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14515 assert(param->unreferenced);
14516 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14522 /* duplicate a context */
14525 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14527 PERL_CONTEXT *ncxs;
14529 PERL_ARGS_ASSERT_CX_DUP;
14532 return (PERL_CONTEXT*)NULL;
14534 /* look for it in the table first */
14535 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14539 /* create anew and remember what it is */
14540 Newx(ncxs, max + 1, PERL_CONTEXT);
14541 ptr_table_store(PL_ptr_table, cxs, ncxs);
14542 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14545 PERL_CONTEXT * const ncx = &ncxs[ix];
14546 if (CxTYPE(ncx) == CXt_SUBST) {
14547 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14550 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14551 switch (CxTYPE(ncx)) {
14553 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14554 if(CxHASARGS(ncx)){
14555 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14557 ncx->blk_sub.savearray = NULL;
14559 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14560 ncx->blk_sub.prevcomppad);
14563 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14565 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14566 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14567 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14568 /* XXX what do do with cur_top_env ???? */
14570 case CXt_LOOP_LAZYSV:
14571 ncx->blk_loop.state_u.lazysv.end
14572 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14573 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14574 duplication code instead.
14575 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14576 actually being the same function, and (2) order
14577 equivalence of the two unions.
14578 We can assert the later [but only at run time :-(] */
14579 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14580 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14583 ncx->blk_loop.state_u.ary.ary
14584 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14586 case CXt_LOOP_LIST:
14587 case CXt_LOOP_LAZYIV:
14588 /* code common to all 'for' CXt_LOOP_* types */
14589 ncx->blk_loop.itersave =
14590 sv_dup_inc(ncx->blk_loop.itersave, param);
14591 if (CxPADLOOP(ncx)) {
14592 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14593 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14594 ncx->blk_loop.oldcomppad =
14595 (PAD*)ptr_table_fetch(PL_ptr_table,
14596 ncx->blk_loop.oldcomppad);
14597 ncx->blk_loop.itervar_u.svp =
14598 &CX_CURPAD_SV(ncx->blk_loop, off);
14601 /* this copies the GV if CXp_FOR_GV, or the SV for an
14602 * alias (for \$x (...)) - relies on gv_dup being the
14603 * same as sv_dup */
14604 ncx->blk_loop.itervar_u.gv
14605 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14609 case CXt_LOOP_PLAIN:
14612 ncx->blk_format.prevcomppad =
14613 (PAD*)ptr_table_fetch(PL_ptr_table,
14614 ncx->blk_format.prevcomppad);
14615 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14616 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14617 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14621 ncx->blk_givwhen.defsv_save =
14622 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14635 /* duplicate a stack info structure */
14638 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14642 PERL_ARGS_ASSERT_SI_DUP;
14645 return (PERL_SI*)NULL;
14647 /* look for it in the table first */
14648 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14652 /* create anew and remember what it is */
14653 Newx(nsi, 1, PERL_SI);
14654 ptr_table_store(PL_ptr_table, si, nsi);
14656 nsi->si_stack = av_dup_inc(si->si_stack, param);
14657 nsi->si_cxix = si->si_cxix;
14658 nsi->si_cxmax = si->si_cxmax;
14659 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14660 nsi->si_type = si->si_type;
14661 nsi->si_prev = si_dup(si->si_prev, param);
14662 nsi->si_next = si_dup(si->si_next, param);
14663 nsi->si_markoff = si->si_markoff;
14664 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14665 nsi->si_stack_hwm = 0;
14671 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14672 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14673 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14674 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14675 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14676 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14677 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14678 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14679 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14680 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14681 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14682 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14683 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14684 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14685 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14686 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14689 #define pv_dup_inc(p) SAVEPV(p)
14690 #define pv_dup(p) SAVEPV(p)
14691 #define svp_dup_inc(p,pp) any_dup(p,pp)
14693 /* map any object to the new equivent - either something in the
14694 * ptr table, or something in the interpreter structure
14698 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14702 PERL_ARGS_ASSERT_ANY_DUP;
14705 return (void*)NULL;
14707 /* look for it in the table first */
14708 ret = ptr_table_fetch(PL_ptr_table, v);
14712 /* see if it is part of the interpreter structure */
14713 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14714 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14722 /* duplicate the save stack */
14725 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14728 ANY * const ss = proto_perl->Isavestack;
14729 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14730 I32 ix = proto_perl->Isavestack_ix;
14743 void (*dptr) (void*);
14744 void (*dxptr) (pTHX_ void*);
14746 PERL_ARGS_ASSERT_SS_DUP;
14748 Newx(nss, max, ANY);
14751 const UV uv = POPUV(ss,ix);
14752 const U8 type = (U8)uv & SAVE_MASK;
14754 TOPUV(nss,ix) = uv;
14756 case SAVEt_CLEARSV:
14757 case SAVEt_CLEARPADRANGE:
14759 case SAVEt_HELEM: /* hash element */
14760 case SAVEt_SV: /* scalar reference */
14761 sv = (const SV *)POPPTR(ss,ix);
14762 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14764 case SAVEt_ITEM: /* normal string */
14765 case SAVEt_GVSV: /* scalar slot in GV */
14766 sv = (const SV *)POPPTR(ss,ix);
14767 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14768 if (type == SAVEt_SV)
14772 case SAVEt_MORTALIZESV:
14773 case SAVEt_READONLY_OFF:
14774 sv = (const SV *)POPPTR(ss,ix);
14775 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14777 case SAVEt_FREEPADNAME:
14778 ptr = POPPTR(ss,ix);
14779 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14780 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14782 case SAVEt_SHARED_PVREF: /* char* in shared space */
14783 c = (char*)POPPTR(ss,ix);
14784 TOPPTR(nss,ix) = savesharedpv(c);
14785 ptr = POPPTR(ss,ix);
14786 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14788 case SAVEt_GENERIC_SVREF: /* generic sv */
14789 case SAVEt_SVREF: /* scalar reference */
14790 sv = (const SV *)POPPTR(ss,ix);
14791 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14792 if (type == SAVEt_SVREF)
14793 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14794 ptr = POPPTR(ss,ix);
14795 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14797 case SAVEt_GVSLOT: /* any slot in GV */
14798 sv = (const SV *)POPPTR(ss,ix);
14799 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14800 ptr = POPPTR(ss,ix);
14801 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14802 sv = (const SV *)POPPTR(ss,ix);
14803 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14805 case SAVEt_HV: /* hash reference */
14806 case SAVEt_AV: /* array reference */
14807 sv = (const SV *) POPPTR(ss,ix);
14808 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14810 case SAVEt_COMPPAD:
14812 sv = (const SV *) POPPTR(ss,ix);
14813 TOPPTR(nss,ix) = sv_dup(sv, param);
14815 case SAVEt_INT: /* int reference */
14816 ptr = POPPTR(ss,ix);
14817 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14818 intval = (int)POPINT(ss,ix);
14819 TOPINT(nss,ix) = intval;
14821 case SAVEt_LONG: /* long reference */
14822 ptr = POPPTR(ss,ix);
14823 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14824 longval = (long)POPLONG(ss,ix);
14825 TOPLONG(nss,ix) = longval;
14827 case SAVEt_I32: /* I32 reference */
14828 ptr = POPPTR(ss,ix);
14829 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14831 TOPINT(nss,ix) = i;
14833 case SAVEt_IV: /* IV reference */
14834 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14835 ptr = POPPTR(ss,ix);
14836 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14838 TOPIV(nss,ix) = iv;
14840 case SAVEt_TMPSFLOOR:
14842 TOPIV(nss,ix) = iv;
14844 case SAVEt_HPTR: /* HV* reference */
14845 case SAVEt_APTR: /* AV* reference */
14846 case SAVEt_SPTR: /* SV* reference */
14847 ptr = POPPTR(ss,ix);
14848 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14849 sv = (const SV *)POPPTR(ss,ix);
14850 TOPPTR(nss,ix) = sv_dup(sv, param);
14852 case SAVEt_VPTR: /* random* reference */
14853 ptr = POPPTR(ss,ix);
14854 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14856 case SAVEt_INT_SMALL:
14857 case SAVEt_I32_SMALL:
14858 case SAVEt_I16: /* I16 reference */
14859 case SAVEt_I8: /* I8 reference */
14861 ptr = POPPTR(ss,ix);
14862 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14864 case SAVEt_GENERIC_PVREF: /* generic char* */
14865 case SAVEt_PPTR: /* char* reference */
14866 ptr = POPPTR(ss,ix);
14867 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14868 c = (char*)POPPTR(ss,ix);
14869 TOPPTR(nss,ix) = pv_dup(c);
14871 case SAVEt_GP: /* scalar reference */
14872 gp = (GP*)POPPTR(ss,ix);
14873 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14874 (void)GpREFCNT_inc(gp);
14875 gv = (const GV *)POPPTR(ss,ix);
14876 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14879 ptr = POPPTR(ss,ix);
14880 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14881 /* these are assumed to be refcounted properly */
14883 switch (((OP*)ptr)->op_type) {
14885 case OP_LEAVESUBLV:
14889 case OP_LEAVEWRITE:
14890 TOPPTR(nss,ix) = ptr;
14893 (void) OpREFCNT_inc(o);
14897 TOPPTR(nss,ix) = NULL;
14902 TOPPTR(nss,ix) = NULL;
14904 case SAVEt_FREECOPHH:
14905 ptr = POPPTR(ss,ix);
14906 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14908 case SAVEt_ADELETE:
14909 av = (const AV *)POPPTR(ss,ix);
14910 TOPPTR(nss,ix) = av_dup_inc(av, param);
14912 TOPINT(nss,ix) = i;
14915 hv = (const HV *)POPPTR(ss,ix);
14916 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14918 TOPINT(nss,ix) = i;
14921 c = (char*)POPPTR(ss,ix);
14922 TOPPTR(nss,ix) = pv_dup_inc(c);
14924 case SAVEt_STACK_POS: /* Position on Perl stack */
14926 TOPINT(nss,ix) = i;
14928 case SAVEt_DESTRUCTOR:
14929 ptr = POPPTR(ss,ix);
14930 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14931 dptr = POPDPTR(ss,ix);
14932 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14933 any_dup(FPTR2DPTR(void *, dptr),
14936 case SAVEt_DESTRUCTOR_X:
14937 ptr = POPPTR(ss,ix);
14938 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14939 dxptr = POPDXPTR(ss,ix);
14940 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14941 any_dup(FPTR2DPTR(void *, dxptr),
14944 case SAVEt_REGCONTEXT:
14946 ix -= uv >> SAVE_TIGHT_SHIFT;
14948 case SAVEt_AELEM: /* array element */
14949 sv = (const SV *)POPPTR(ss,ix);
14950 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14952 TOPIV(nss,ix) = iv;
14953 av = (const AV *)POPPTR(ss,ix);
14954 TOPPTR(nss,ix) = av_dup_inc(av, param);
14957 ptr = POPPTR(ss,ix);
14958 TOPPTR(nss,ix) = ptr;
14961 ptr = POPPTR(ss,ix);
14962 ptr = cophh_copy((COPHH*)ptr);
14963 TOPPTR(nss,ix) = ptr;
14965 TOPINT(nss,ix) = i;
14966 if (i & HINT_LOCALIZE_HH) {
14967 hv = (const HV *)POPPTR(ss,ix);
14968 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14971 case SAVEt_PADSV_AND_MORTALIZE:
14972 longval = (long)POPLONG(ss,ix);
14973 TOPLONG(nss,ix) = longval;
14974 ptr = POPPTR(ss,ix);
14975 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14976 sv = (const SV *)POPPTR(ss,ix);
14977 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14979 case SAVEt_SET_SVFLAGS:
14981 TOPINT(nss,ix) = i;
14983 TOPINT(nss,ix) = i;
14984 sv = (const SV *)POPPTR(ss,ix);
14985 TOPPTR(nss,ix) = sv_dup(sv, param);
14987 case SAVEt_COMPILE_WARNINGS:
14988 ptr = POPPTR(ss,ix);
14989 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14992 ptr = POPPTR(ss,ix);
14993 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14997 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
15005 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15006 * flag to the result. This is done for each stash before cloning starts,
15007 * so we know which stashes want their objects cloned */
15010 do_mark_cloneable_stash(pTHX_ SV *const sv)
15012 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15014 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15015 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15016 if (cloner && GvCV(cloner)) {
15023 mXPUSHs(newSVhek(hvname));
15025 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15032 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15040 =for apidoc perl_clone
15042 Create and return a new interpreter by cloning the current one.
15044 C<perl_clone> takes these flags as parameters:
15046 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15047 without it we only clone the data and zero the stacks,
15048 with it we copy the stacks and the new perl interpreter is
15049 ready to run at the exact same point as the previous one.
15050 The pseudo-fork code uses C<COPY_STACKS> while the
15051 threads->create doesn't.
15053 C<CLONEf_KEEP_PTR_TABLE> -
15054 C<perl_clone> keeps a ptr_table with the pointer of the old
15055 variable as a key and the new variable as a value,
15056 this allows it to check if something has been cloned and not
15057 clone it again but rather just use the value and increase the
15058 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15059 the ptr_table using the function
15060 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15061 reason to keep it around is if you want to dup some of your own
15062 variable who are outside the graph perl scans, an example of this
15063 code is in F<threads.xs> create.
15065 C<CLONEf_CLONE_HOST> -
15066 This is a win32 thing, it is ignored on unix, it tells perls
15067 win32host code (which is c++) to clone itself, this is needed on
15068 win32 if you want to run two threads at the same time,
15069 if you just want to do some stuff in a separate perl interpreter
15070 and then throw it away and return to the original one,
15071 you don't need to do anything.
15076 /* XXX the above needs expanding by someone who actually understands it ! */
15077 EXTERN_C PerlInterpreter *
15078 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15081 perl_clone(PerlInterpreter *proto_perl, UV flags)
15084 #ifdef PERL_IMPLICIT_SYS
15086 PERL_ARGS_ASSERT_PERL_CLONE;
15088 /* perlhost.h so we need to call into it
15089 to clone the host, CPerlHost should have a c interface, sky */
15091 #ifndef __amigaos4__
15092 if (flags & CLONEf_CLONE_HOST) {
15093 return perl_clone_host(proto_perl,flags);
15096 return perl_clone_using(proto_perl, flags,
15098 proto_perl->IMemShared,
15099 proto_perl->IMemParse,
15101 proto_perl->IStdIO,
15105 proto_perl->IProc);
15109 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15110 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15111 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15112 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15113 struct IPerlDir* ipD, struct IPerlSock* ipS,
15114 struct IPerlProc* ipP)
15116 /* XXX many of the string copies here can be optimized if they're
15117 * constants; they need to be allocated as common memory and just
15118 * their pointers copied. */
15121 CLONE_PARAMS clone_params;
15122 CLONE_PARAMS* const param = &clone_params;
15124 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15126 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15127 #else /* !PERL_IMPLICIT_SYS */
15129 CLONE_PARAMS clone_params;
15130 CLONE_PARAMS* param = &clone_params;
15131 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15133 PERL_ARGS_ASSERT_PERL_CLONE;
15134 #endif /* PERL_IMPLICIT_SYS */
15136 /* for each stash, determine whether its objects should be cloned */
15137 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15138 PERL_SET_THX(my_perl);
15141 PoisonNew(my_perl, 1, PerlInterpreter);
15144 PL_defstash = NULL; /* may be used by perl malloc() */
15147 PL_scopestack_name = 0;
15149 PL_savestack_ix = 0;
15150 PL_savestack_max = -1;
15151 PL_sig_pending = 0;
15153 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15154 Zero(&PL_padname_undef, 1, PADNAME);
15155 Zero(&PL_padname_const, 1, PADNAME);
15156 # ifdef DEBUG_LEAKING_SCALARS
15157 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15159 # ifdef PERL_TRACE_OPS
15160 Zero(PL_op_exec_cnt, OP_max+2, UV);
15162 #else /* !DEBUGGING */
15163 Zero(my_perl, 1, PerlInterpreter);
15164 #endif /* DEBUGGING */
15166 #ifdef PERL_IMPLICIT_SYS
15167 /* host pointers */
15169 PL_MemShared = ipMS;
15170 PL_MemParse = ipMP;
15177 #endif /* PERL_IMPLICIT_SYS */
15180 param->flags = flags;
15181 /* Nothing in the core code uses this, but we make it available to
15182 extensions (using mg_dup). */
15183 param->proto_perl = proto_perl;
15184 /* Likely nothing will use this, but it is initialised to be consistent
15185 with Perl_clone_params_new(). */
15186 param->new_perl = my_perl;
15187 param->unreferenced = NULL;
15190 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15192 PL_body_arenas = NULL;
15193 Zero(&PL_body_roots, 1, PL_body_roots);
15197 PL_sv_arenaroot = NULL;
15199 PL_debug = proto_perl->Idebug;
15201 /* dbargs array probably holds garbage */
15204 PL_compiling = proto_perl->Icompiling;
15206 /* pseudo environmental stuff */
15207 PL_origargc = proto_perl->Iorigargc;
15208 PL_origargv = proto_perl->Iorigargv;
15210 #ifndef NO_TAINT_SUPPORT
15211 /* Set tainting stuff before PerlIO_debug can possibly get called */
15212 PL_tainting = proto_perl->Itainting;
15213 PL_taint_warn = proto_perl->Itaint_warn;
15215 PL_tainting = FALSE;
15216 PL_taint_warn = FALSE;
15219 PL_minus_c = proto_perl->Iminus_c;
15221 PL_localpatches = proto_perl->Ilocalpatches;
15222 PL_splitstr = proto_perl->Isplitstr;
15223 PL_minus_n = proto_perl->Iminus_n;
15224 PL_minus_p = proto_perl->Iminus_p;
15225 PL_minus_l = proto_perl->Iminus_l;
15226 PL_minus_a = proto_perl->Iminus_a;
15227 PL_minus_E = proto_perl->Iminus_E;
15228 PL_minus_F = proto_perl->Iminus_F;
15229 PL_doswitches = proto_perl->Idoswitches;
15230 PL_dowarn = proto_perl->Idowarn;
15231 #ifdef PERL_SAWAMPERSAND
15232 PL_sawampersand = proto_perl->Isawampersand;
15234 PL_unsafe = proto_perl->Iunsafe;
15235 PL_perldb = proto_perl->Iperldb;
15236 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15237 PL_exit_flags = proto_perl->Iexit_flags;
15239 /* XXX time(&PL_basetime) when asked for? */
15240 PL_basetime = proto_perl->Ibasetime;
15242 PL_maxsysfd = proto_perl->Imaxsysfd;
15243 PL_statusvalue = proto_perl->Istatusvalue;
15245 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15247 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15250 /* RE engine related */
15251 PL_regmatch_slab = NULL;
15252 PL_reg_curpm = NULL;
15254 PL_sub_generation = proto_perl->Isub_generation;
15256 /* funky return mechanisms */
15257 PL_forkprocess = proto_perl->Iforkprocess;
15259 /* internal state */
15260 PL_main_start = proto_perl->Imain_start;
15261 PL_eval_root = proto_perl->Ieval_root;
15262 PL_eval_start = proto_perl->Ieval_start;
15264 PL_filemode = proto_perl->Ifilemode;
15265 PL_lastfd = proto_perl->Ilastfd;
15266 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15267 PL_gensym = proto_perl->Igensym;
15269 PL_laststatval = proto_perl->Ilaststatval;
15270 PL_laststype = proto_perl->Ilaststype;
15273 PL_profiledata = NULL;
15275 PL_generation = proto_perl->Igeneration;
15277 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15278 PL_in_clean_all = proto_perl->Iin_clean_all;
15280 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15281 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15282 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15283 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15284 PL_nomemok = proto_perl->Inomemok;
15285 PL_an = proto_perl->Ian;
15286 PL_evalseq = proto_perl->Ievalseq;
15287 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15288 PL_origalen = proto_perl->Iorigalen;
15290 PL_sighandlerp = proto_perl->Isighandlerp;
15292 PL_runops = proto_perl->Irunops;
15294 PL_subline = proto_perl->Isubline;
15296 PL_cv_has_eval = proto_perl->Icv_has_eval;
15299 PL_cryptseen = proto_perl->Icryptseen;
15302 #ifdef USE_LOCALE_COLLATE
15303 PL_collation_ix = proto_perl->Icollation_ix;
15304 PL_collation_standard = proto_perl->Icollation_standard;
15305 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15306 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15307 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15308 #endif /* USE_LOCALE_COLLATE */
15310 #ifdef USE_LOCALE_NUMERIC
15311 PL_numeric_standard = proto_perl->Inumeric_standard;
15312 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15313 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15314 #endif /* !USE_LOCALE_NUMERIC */
15316 /* Did the locale setup indicate UTF-8? */
15317 PL_utf8locale = proto_perl->Iutf8locale;
15318 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15319 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15320 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15321 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15322 PL_lc_numeric_mutex_depth = 0;
15324 /* Unicode features (see perlrun/-C) */
15325 PL_unicode = proto_perl->Iunicode;
15327 /* Pre-5.8 signals control */
15328 PL_signals = proto_perl->Isignals;
15330 /* times() ticks per second */
15331 PL_clocktick = proto_perl->Iclocktick;
15333 /* Recursion stopper for PerlIO_find_layer */
15334 PL_in_load_module = proto_perl->Iin_load_module;
15336 /* sort() routine */
15337 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15339 /* Not really needed/useful since the reenrant_retint is "volatile",
15340 * but do it for consistency's sake. */
15341 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15343 /* Hooks to shared SVs and locks. */
15344 PL_sharehook = proto_perl->Isharehook;
15345 PL_lockhook = proto_perl->Ilockhook;
15346 PL_unlockhook = proto_perl->Iunlockhook;
15347 PL_threadhook = proto_perl->Ithreadhook;
15348 PL_destroyhook = proto_perl->Idestroyhook;
15349 PL_signalhook = proto_perl->Isignalhook;
15351 PL_globhook = proto_perl->Iglobhook;
15354 PL_last_swash_hv = NULL; /* reinits on demand */
15355 PL_last_swash_klen = 0;
15356 PL_last_swash_key[0]= '\0';
15357 PL_last_swash_tmps = (U8*)NULL;
15358 PL_last_swash_slen = 0;
15360 PL_srand_called = proto_perl->Isrand_called;
15361 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15363 if (flags & CLONEf_COPY_STACKS) {
15364 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15365 PL_tmps_ix = proto_perl->Itmps_ix;
15366 PL_tmps_max = proto_perl->Itmps_max;
15367 PL_tmps_floor = proto_perl->Itmps_floor;
15369 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15370 * NOTE: unlike the others! */
15371 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15372 PL_scopestack_max = proto_perl->Iscopestack_max;
15374 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15375 * NOTE: unlike the others! */
15376 PL_savestack_ix = proto_perl->Isavestack_ix;
15377 PL_savestack_max = proto_perl->Isavestack_max;
15380 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15381 PL_top_env = &PL_start_env;
15383 PL_op = proto_perl->Iop;
15386 PL_Xpv = (XPV*)NULL;
15387 my_perl->Ina = proto_perl->Ina;
15389 PL_statcache = proto_perl->Istatcache;
15391 #ifndef NO_TAINT_SUPPORT
15392 PL_tainted = proto_perl->Itainted;
15394 PL_tainted = FALSE;
15396 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15398 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15400 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15401 PL_restartop = proto_perl->Irestartop;
15402 PL_in_eval = proto_perl->Iin_eval;
15403 PL_delaymagic = proto_perl->Idelaymagic;
15404 PL_phase = proto_perl->Iphase;
15405 PL_localizing = proto_perl->Ilocalizing;
15407 PL_hv_fetch_ent_mh = NULL;
15408 PL_modcount = proto_perl->Imodcount;
15409 PL_lastgotoprobe = NULL;
15410 PL_dumpindent = proto_perl->Idumpindent;
15412 PL_efloatbuf = NULL; /* reinits on demand */
15413 PL_efloatsize = 0; /* reinits on demand */
15417 PL_colorset = 0; /* reinits PL_colors[] */
15418 /*PL_colors[6] = {0,0,0,0,0,0};*/
15420 /* Pluggable optimizer */
15421 PL_peepp = proto_perl->Ipeepp;
15422 PL_rpeepp = proto_perl->Irpeepp;
15423 /* op_free() hook */
15424 PL_opfreehook = proto_perl->Iopfreehook;
15426 #ifdef USE_REENTRANT_API
15427 /* XXX: things like -Dm will segfault here in perlio, but doing
15428 * PERL_SET_CONTEXT(proto_perl);
15429 * breaks too many other things
15431 Perl_reentrant_init(aTHX);
15434 /* create SV map for pointer relocation */
15435 PL_ptr_table = ptr_table_new();
15437 /* initialize these special pointers as early as possible */
15439 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15440 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15441 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15442 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15443 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15444 &PL_padname_const);
15446 /* create (a non-shared!) shared string table */
15447 PL_strtab = newHV();
15448 HvSHAREKEYS_off(PL_strtab);
15449 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15450 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15452 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15454 /* This PV will be free'd special way so must set it same way op.c does */
15455 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15456 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15458 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15459 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15460 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15461 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15463 param->stashes = newAV(); /* Setup array of objects to call clone on */
15464 /* This makes no difference to the implementation, as it always pushes
15465 and shifts pointers to other SVs without changing their reference
15466 count, with the array becoming empty before it is freed. However, it
15467 makes it conceptually clear what is going on, and will avoid some
15468 work inside av.c, filling slots between AvFILL() and AvMAX() with
15469 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15470 AvREAL_off(param->stashes);
15472 if (!(flags & CLONEf_COPY_STACKS)) {
15473 param->unreferenced = newAV();
15476 #ifdef PERLIO_LAYERS
15477 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15478 PerlIO_clone(aTHX_ proto_perl, param);
15481 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15482 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15483 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15484 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15485 PL_xsubfilename = proto_perl->Ixsubfilename;
15486 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15487 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15490 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15491 PL_inplace = SAVEPV(proto_perl->Iinplace);
15492 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15494 /* magical thingies */
15496 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15497 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15498 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15501 /* Clone the regex array */
15502 /* ORANGE FIXME for plugins, probably in the SV dup code.
15503 newSViv(PTR2IV(CALLREGDUPE(
15504 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15506 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15507 PL_regex_pad = AvARRAY(PL_regex_padav);
15509 PL_stashpadmax = proto_perl->Istashpadmax;
15510 PL_stashpadix = proto_perl->Istashpadix ;
15511 Newx(PL_stashpad, PL_stashpadmax, HV *);
15514 for (; o < PL_stashpadmax; ++o)
15515 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15518 /* shortcuts to various I/O objects */
15519 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15520 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15521 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15522 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15523 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15524 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15525 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15527 /* shortcuts to regexp stuff */
15528 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15530 /* shortcuts to misc objects */
15531 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15533 /* shortcuts to debugging objects */
15534 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15535 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15536 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15537 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15538 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15539 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15540 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15542 /* symbol tables */
15543 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15544 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15545 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15546 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15547 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15549 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15550 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15551 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15552 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15553 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15554 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15555 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15556 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15557 PL_savebegin = proto_perl->Isavebegin;
15559 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15561 /* subprocess state */
15562 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15564 if (proto_perl->Iop_mask)
15565 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15568 /* PL_asserting = proto_perl->Iasserting; */
15570 /* current interpreter roots */
15571 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15573 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15576 /* runtime control stuff */
15577 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15579 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15581 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15583 /* interpreter atexit processing */
15584 PL_exitlistlen = proto_perl->Iexitlistlen;
15585 if (PL_exitlistlen) {
15586 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15587 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15590 PL_exitlist = (PerlExitListEntry*)NULL;
15592 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15593 if (PL_my_cxt_size) {
15594 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15595 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15598 PL_my_cxt_list = (void**)NULL;
15600 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15601 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15602 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15603 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15605 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15607 PAD_CLONE_VARS(proto_perl, param);
15609 #ifdef HAVE_INTERP_INTERN
15610 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15613 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15615 #ifdef PERL_USES_PL_PIDSTATUS
15616 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15618 PL_osname = SAVEPV(proto_perl->Iosname);
15619 PL_parser = parser_dup(proto_perl->Iparser, param);
15621 /* XXX this only works if the saved cop has already been cloned */
15622 if (proto_perl->Iparser) {
15623 PL_parser->saved_curcop = (COP*)any_dup(
15624 proto_perl->Iparser->saved_curcop,
15628 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15630 #if defined(USE_POSIX_2008_LOCALE) \
15631 && defined(USE_THREAD_SAFE_LOCALE) \
15632 && ! defined(HAS_QUERYLOCALE)
15633 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15634 PL_curlocales[i] = savepv("."); /* An illegal value */
15637 #ifdef USE_LOCALE_CTYPE
15638 /* Should we warn if uses locale? */
15639 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15642 #ifdef USE_LOCALE_COLLATE
15643 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15644 #endif /* USE_LOCALE_COLLATE */
15646 #ifdef USE_LOCALE_NUMERIC
15647 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15648 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15650 # if defined(HAS_POSIX_2008_LOCALE)
15651 PL_underlying_numeric_obj = NULL;
15653 #endif /* !USE_LOCALE_NUMERIC */
15655 PL_langinfo_buf = NULL;
15656 PL_langinfo_bufsize = 0;
15658 PL_setlocale_buf = NULL;
15659 PL_setlocale_bufsize = 0;
15661 /* utf8 character class swashes */
15662 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15664 if (proto_perl->Ipsig_pend) {
15665 Newxz(PL_psig_pend, SIG_SIZE, int);
15668 PL_psig_pend = (int*)NULL;
15671 if (proto_perl->Ipsig_name) {
15672 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15673 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15675 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15678 PL_psig_ptr = (SV**)NULL;
15679 PL_psig_name = (SV**)NULL;
15682 if (flags & CLONEf_COPY_STACKS) {
15683 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15684 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15685 PL_tmps_ix+1, param);
15687 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15688 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15689 Newx(PL_markstack, i, I32);
15690 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15691 - proto_perl->Imarkstack);
15692 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15693 - proto_perl->Imarkstack);
15694 Copy(proto_perl->Imarkstack, PL_markstack,
15695 PL_markstack_ptr - PL_markstack + 1, I32);
15697 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15698 * NOTE: unlike the others! */
15699 Newx(PL_scopestack, PL_scopestack_max, I32);
15700 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15703 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15704 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15706 /* reset stack AV to correct length before its duped via
15707 * PL_curstackinfo */
15708 AvFILLp(proto_perl->Icurstack) =
15709 proto_perl->Istack_sp - proto_perl->Istack_base;
15711 /* NOTE: si_dup() looks at PL_markstack */
15712 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15714 /* PL_curstack = PL_curstackinfo->si_stack; */
15715 PL_curstack = av_dup(proto_perl->Icurstack, param);
15716 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15718 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15719 PL_stack_base = AvARRAY(PL_curstack);
15720 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15721 - proto_perl->Istack_base);
15722 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15724 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15725 PL_savestack = ss_dup(proto_perl, param);
15729 ENTER; /* perl_destruct() wants to LEAVE; */
15732 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15733 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15735 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15736 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15737 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15738 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15739 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15740 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15742 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15744 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15745 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15746 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15748 PL_stashcache = newHV();
15750 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15751 proto_perl->Iwatchaddr);
15752 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15753 if (PL_debug && PL_watchaddr) {
15754 PerlIO_printf(Perl_debug_log,
15755 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15756 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15757 PTR2UV(PL_watchok));
15760 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15761 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15763 /* Call the ->CLONE method, if it exists, for each of the stashes
15764 identified by sv_dup() above.
15766 while(av_tindex(param->stashes) != -1) {
15767 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15768 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15769 if (cloner && GvCV(cloner)) {
15774 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15776 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15782 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15783 ptr_table_free(PL_ptr_table);
15784 PL_ptr_table = NULL;
15787 if (!(flags & CLONEf_COPY_STACKS)) {
15788 unreferenced_to_tmp_stack(param->unreferenced);
15791 SvREFCNT_dec(param->stashes);
15793 /* orphaned? eg threads->new inside BEGIN or use */
15794 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15795 SvREFCNT_inc_simple_void(PL_compcv);
15796 SAVEFREESV(PL_compcv);
15803 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15805 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15807 if (AvFILLp(unreferenced) > -1) {
15808 SV **svp = AvARRAY(unreferenced);
15809 SV **const last = svp + AvFILLp(unreferenced);
15813 if (SvREFCNT(*svp) == 1)
15815 } while (++svp <= last);
15817 EXTEND_MORTAL(count);
15818 svp = AvARRAY(unreferenced);
15821 if (SvREFCNT(*svp) == 1) {
15822 /* Our reference is the only one to this SV. This means that
15823 in this thread, the scalar effectively has a 0 reference.
15824 That doesn't work (cleanup never happens), so donate our
15825 reference to it onto the save stack. */
15826 PL_tmps_stack[++PL_tmps_ix] = *svp;
15828 /* As an optimisation, because we are already walking the
15829 entire array, instead of above doing either
15830 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15831 release our reference to the scalar, so that at the end of
15832 the array owns zero references to the scalars it happens to
15833 point to. We are effectively converting the array from
15834 AvREAL() on to AvREAL() off. This saves the av_clear()
15835 (triggered by the SvREFCNT_dec(unreferenced) below) from
15836 walking the array a second time. */
15837 SvREFCNT_dec(*svp);
15840 } while (++svp <= last);
15841 AvREAL_off(unreferenced);
15843 SvREFCNT_dec_NN(unreferenced);
15847 Perl_clone_params_del(CLONE_PARAMS *param)
15849 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15851 PerlInterpreter *const to = param->new_perl;
15853 PerlInterpreter *const was = PERL_GET_THX;
15855 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15861 SvREFCNT_dec(param->stashes);
15862 if (param->unreferenced)
15863 unreferenced_to_tmp_stack(param->unreferenced);
15873 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15876 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15877 does a dTHX; to get the context from thread local storage.
15878 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15879 a version that passes in my_perl. */
15880 PerlInterpreter *const was = PERL_GET_THX;
15881 CLONE_PARAMS *param;
15883 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15889 /* Given that we've set the context, we can do this unshared. */
15890 Newx(param, 1, CLONE_PARAMS);
15893 param->proto_perl = from;
15894 param->new_perl = to;
15895 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15896 AvREAL_off(param->stashes);
15897 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15905 #endif /* USE_ITHREADS */
15908 Perl_init_constants(pTHX)
15912 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15913 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15914 SvANY(&PL_sv_undef) = NULL;
15916 SvANY(&PL_sv_no) = new_XPVNV();
15917 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15918 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15919 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15922 SvANY(&PL_sv_yes) = new_XPVNV();
15923 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15924 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15925 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15928 SvANY(&PL_sv_zero) = new_XPVNV();
15929 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15930 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15931 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15935 SvPV_set(&PL_sv_no, (char*)PL_No);
15936 SvCUR_set(&PL_sv_no, 0);
15937 SvLEN_set(&PL_sv_no, 0);
15938 SvIV_set(&PL_sv_no, 0);
15939 SvNV_set(&PL_sv_no, 0);
15941 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15942 SvCUR_set(&PL_sv_yes, 1);
15943 SvLEN_set(&PL_sv_yes, 0);
15944 SvIV_set(&PL_sv_yes, 1);
15945 SvNV_set(&PL_sv_yes, 1);
15947 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15948 SvCUR_set(&PL_sv_zero, 1);
15949 SvLEN_set(&PL_sv_zero, 0);
15950 SvIV_set(&PL_sv_zero, 0);
15951 SvNV_set(&PL_sv_zero, 0);
15953 PadnamePV(&PL_padname_const) = (char *)PL_No;
15955 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15956 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15957 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15958 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15960 assert(SvIMMORTAL(&PL_sv_yes));
15961 assert(SvIMMORTAL(&PL_sv_undef));
15962 assert(SvIMMORTAL(&PL_sv_no));
15963 assert(SvIMMORTAL(&PL_sv_zero));
15965 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
15966 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
15967 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
15968 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
15970 assert( SvTRUE_nomg_NN(&PL_sv_yes));
15971 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
15972 assert(!SvTRUE_nomg_NN(&PL_sv_no));
15973 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
15977 =head1 Unicode Support
15979 =for apidoc sv_recode_to_utf8
15981 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15982 of C<sv> is assumed to be octets in that encoding, and C<sv>
15983 will be converted into Unicode (and UTF-8).
15985 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15986 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15987 an C<Encode::XS> Encoding object, bad things will happen.
15988 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15990 The PV of C<sv> is returned.
15995 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15997 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15999 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
16008 if (SvPADTMP(nsv)) {
16009 nsv = sv_newmortal();
16010 SvSetSV_nosteal(nsv, sv);
16019 Passing sv_yes is wrong - it needs to be or'ed set of constants
16020 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16021 remove converted chars from source.
16023 Both will default the value - let them.
16025 XPUSHs(&PL_sv_yes);
16028 call_method("decode", G_SCALAR);
16032 s = SvPV_const(uni, len);
16033 if (s != SvPVX_const(sv)) {
16034 SvGROW(sv, len + 1);
16035 Move(s, SvPVX(sv), len + 1, char);
16036 SvCUR_set(sv, len);
16041 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16042 /* clear pos and any utf8 cache */
16043 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16046 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16047 magic_setutf8(sv,mg); /* clear UTF8 cache */
16052 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16056 =for apidoc sv_cat_decode
16058 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16059 assumed to be octets in that encoding and decoding the input starts
16060 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16061 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16062 when the string C<tstr> appears in decoding output or the input ends on
16063 the PV of C<ssv>. The value which C<offset> points will be modified
16064 to the last input position on C<ssv>.
16066 Returns TRUE if the terminator was found, else returns FALSE.
16071 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16072 SV *ssv, int *offset, char *tstr, int tlen)
16076 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16078 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16089 offsv = newSViv(*offset);
16091 mPUSHp(tstr, tlen);
16093 call_method("cat_decode", G_SCALAR);
16095 ret = SvTRUE(TOPs);
16096 *offset = SvIV(offsv);
16102 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16107 /* ---------------------------------------------------------------------
16109 * support functions for report_uninit()
16112 /* the maxiumum size of array or hash where we will scan looking
16113 * for the undefined element that triggered the warning */
16115 #define FUV_MAX_SEARCH_SIZE 1000
16117 /* Look for an entry in the hash whose value has the same SV as val;
16118 * If so, return a mortal copy of the key. */
16121 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16127 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16129 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16130 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16133 array = HvARRAY(hv);
16135 for (i=HvMAX(hv); i>=0; i--) {
16137 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16138 if (HeVAL(entry) != val)
16140 if ( HeVAL(entry) == &PL_sv_undef ||
16141 HeVAL(entry) == &PL_sv_placeholder)
16145 if (HeKLEN(entry) == HEf_SVKEY)
16146 return sv_mortalcopy(HeKEY_sv(entry));
16147 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16153 /* Look for an entry in the array whose value has the same SV as val;
16154 * If so, return the index, otherwise return -1. */
16157 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16159 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16161 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16162 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16165 if (val != &PL_sv_undef) {
16166 SV ** const svp = AvARRAY(av);
16169 for (i=AvFILLp(av); i>=0; i--)
16176 /* varname(): return the name of a variable, optionally with a subscript.
16177 * If gv is non-zero, use the name of that global, along with gvtype (one
16178 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16179 * targ. Depending on the value of the subscript_type flag, return:
16182 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16183 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16184 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16185 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16188 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16189 const SV *const keyname, SSize_t aindex, int subscript_type)
16192 SV * const name = sv_newmortal();
16193 if (gv && isGV(gv)) {
16195 buffer[0] = gvtype;
16198 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16200 gv_fullname4(name, gv, buffer, 0);
16202 if ((unsigned int)SvPVX(name)[1] <= 26) {
16204 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16206 /* Swap the 1 unprintable control character for the 2 byte pretty
16207 version - ie substr($name, 1, 1) = $buffer; */
16208 sv_insert(name, 1, 1, buffer, 2);
16212 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16215 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16217 if (!cv || !CvPADLIST(cv))
16219 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16220 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16224 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16225 SV * const sv = newSV(0);
16227 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16229 *SvPVX(name) = '$';
16230 Perl_sv_catpvf(aTHX_ name, "{%s}",
16231 pv_pretty(sv, pv, len, 32, NULL, NULL,
16232 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16233 SvREFCNT_dec_NN(sv);
16235 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16236 *SvPVX(name) = '$';
16237 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16239 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16240 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16241 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16249 =for apidoc find_uninit_var
16251 Find the name of the undefined variable (if any) that caused the operator
16252 to issue a "Use of uninitialized value" warning.
16253 If match is true, only return a name if its value matches C<uninit_sv>.
16254 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16255 warning, then following the direct child of the op may yield an
16256 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16257 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16258 the variable name if we get an exact match.
16259 C<desc_p> points to a string pointer holding the description of the op.
16260 This may be updated if needed.
16262 The name is returned as a mortal SV.
16264 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16265 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16271 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16272 bool match, const char **desc_p)
16277 const OP *o, *o2, *kid;
16279 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16281 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16282 uninit_sv == &PL_sv_placeholder)))
16285 switch (obase->op_type) {
16288 /* undef should care if its args are undef - any warnings
16289 * will be from tied/magic vars */
16297 const bool pad = ( obase->op_type == OP_PADAV
16298 || obase->op_type == OP_PADHV
16299 || obase->op_type == OP_PADRANGE
16302 const bool hash = ( obase->op_type == OP_PADHV
16303 || obase->op_type == OP_RV2HV
16304 || (obase->op_type == OP_PADRANGE
16305 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16309 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16311 if (pad) { /* @lex, %lex */
16312 sv = PAD_SVl(obase->op_targ);
16316 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16317 /* @global, %global */
16318 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16321 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16323 else if (obase == PL_op) /* @{expr}, %{expr} */
16324 return find_uninit_var(cUNOPx(obase)->op_first,
16325 uninit_sv, match, desc_p);
16326 else /* @{expr}, %{expr} as a sub-expression */
16330 /* attempt to find a match within the aggregate */
16332 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16334 subscript_type = FUV_SUBSCRIPT_HASH;
16337 index = find_array_subscript((const AV *)sv, uninit_sv);
16339 subscript_type = FUV_SUBSCRIPT_ARRAY;
16342 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16345 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16346 keysv, index, subscript_type);
16350 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16352 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16353 if (!gv || !GvSTASH(gv))
16355 if (match && (GvSV(gv) != uninit_sv))
16357 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16360 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16363 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16365 return varname(NULL, '$', obase->op_targ,
16366 NULL, 0, FUV_SUBSCRIPT_NONE);
16369 gv = cGVOPx_gv(obase);
16370 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16372 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16374 case OP_AELEMFAST_LEX:
16377 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16378 if (!av || SvRMAGICAL(av))
16380 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16381 if (!svp || *svp != uninit_sv)
16384 return varname(NULL, '$', obase->op_targ,
16385 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16388 gv = cGVOPx_gv(obase);
16393 AV *const av = GvAV(gv);
16394 if (!av || SvRMAGICAL(av))
16396 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16397 if (!svp || *svp != uninit_sv)
16400 return varname(gv, '$', 0,
16401 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16403 NOT_REACHED; /* NOTREACHED */
16406 o = cUNOPx(obase)->op_first;
16407 if (!o || o->op_type != OP_NULL ||
16408 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16410 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16415 bool negate = FALSE;
16417 if (PL_op == obase)
16418 /* $a[uninit_expr] or $h{uninit_expr} */
16419 return find_uninit_var(cBINOPx(obase)->op_last,
16420 uninit_sv, match, desc_p);
16423 o = cBINOPx(obase)->op_first;
16424 kid = cBINOPx(obase)->op_last;
16426 /* get the av or hv, and optionally the gv */
16428 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16429 sv = PAD_SV(o->op_targ);
16431 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16432 && cUNOPo->op_first->op_type == OP_GV)
16434 gv = cGVOPx_gv(cUNOPo->op_first);
16438 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16443 if (kid && kid->op_type == OP_NEGATE) {
16445 kid = cUNOPx(kid)->op_first;
16448 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16449 /* index is constant */
16452 kidsv = newSVpvs_flags("-", SVs_TEMP);
16453 sv_catsv(kidsv, cSVOPx_sv(kid));
16456 kidsv = cSVOPx_sv(kid);
16460 if (obase->op_type == OP_HELEM) {
16461 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16462 if (!he || HeVAL(he) != uninit_sv)
16466 SV * const opsv = cSVOPx_sv(kid);
16467 const IV opsviv = SvIV(opsv);
16468 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16469 negate ? - opsviv : opsviv,
16471 if (!svp || *svp != uninit_sv)
16475 if (obase->op_type == OP_HELEM)
16476 return varname(gv, '%', o->op_targ,
16477 kidsv, 0, FUV_SUBSCRIPT_HASH);
16479 return varname(gv, '@', o->op_targ, NULL,
16480 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16481 FUV_SUBSCRIPT_ARRAY);
16484 /* index is an expression;
16485 * attempt to find a match within the aggregate */
16486 if (obase->op_type == OP_HELEM) {
16487 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16489 return varname(gv, '%', o->op_targ,
16490 keysv, 0, FUV_SUBSCRIPT_HASH);
16493 const SSize_t index
16494 = find_array_subscript((const AV *)sv, uninit_sv);
16496 return varname(gv, '@', o->op_targ,
16497 NULL, index, FUV_SUBSCRIPT_ARRAY);
16502 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16504 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16506 NOT_REACHED; /* NOTREACHED */
16509 case OP_MULTIDEREF: {
16510 /* If we were executing OP_MULTIDEREF when the undef warning
16511 * triggered, then it must be one of the index values within
16512 * that triggered it. If not, then the only possibility is that
16513 * the value retrieved by the last aggregate index might be the
16514 * culprit. For the former, we set PL_multideref_pc each time before
16515 * using an index, so work though the item list until we reach
16516 * that point. For the latter, just work through the entire item
16517 * list; the last aggregate retrieved will be the candidate.
16518 * There is a third rare possibility: something triggered
16519 * magic while fetching an array/hash element. Just display
16520 * nothing in this case.
16523 /* the named aggregate, if any */
16524 PADOFFSET agg_targ = 0;
16526 /* the last-seen index */
16528 PADOFFSET index_targ;
16530 IV index_const_iv = 0; /* init for spurious compiler warn */
16531 SV *index_const_sv;
16532 int depth = 0; /* how many array/hash lookups we've done */
16534 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16535 UNOP_AUX_item *last = NULL;
16536 UV actions = items->uv;
16539 if (PL_op == obase) {
16540 last = PL_multideref_pc;
16541 assert(last >= items && last <= items + items[-1].uv);
16548 switch (actions & MDEREF_ACTION_MASK) {
16550 case MDEREF_reload:
16551 actions = (++items)->uv;
16554 case MDEREF_HV_padhv_helem: /* $lex{...} */
16557 case MDEREF_AV_padav_aelem: /* $lex[...] */
16558 agg_targ = (++items)->pad_offset;
16562 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16565 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16567 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16568 assert(isGV_with_GP(agg_gv));
16571 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16572 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16575 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16576 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16582 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16583 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16586 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16587 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16594 index_const_sv = NULL;
16596 index_type = (actions & MDEREF_INDEX_MASK);
16597 switch (index_type) {
16598 case MDEREF_INDEX_none:
16600 case MDEREF_INDEX_const:
16602 index_const_sv = UNOP_AUX_item_sv(++items)
16604 index_const_iv = (++items)->iv;
16606 case MDEREF_INDEX_padsv:
16607 index_targ = (++items)->pad_offset;
16609 case MDEREF_INDEX_gvsv:
16610 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16611 assert(isGV_with_GP(index_gv));
16615 if (index_type != MDEREF_INDEX_none)
16618 if ( index_type == MDEREF_INDEX_none
16619 || (actions & MDEREF_FLAG_last)
16620 || (last && items >= last)
16624 actions >>= MDEREF_SHIFT;
16627 if (PL_op == obase) {
16628 /* most likely index was undef */
16630 *desc_p = ( (actions & MDEREF_FLAG_last)
16631 && (obase->op_private
16632 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16634 (obase->op_private & OPpMULTIDEREF_EXISTS)
16637 : is_hv ? "hash element" : "array element";
16638 assert(index_type != MDEREF_INDEX_none);
16640 if (GvSV(index_gv) == uninit_sv)
16641 return varname(index_gv, '$', 0, NULL, 0,
16642 FUV_SUBSCRIPT_NONE);
16647 if (PL_curpad[index_targ] == uninit_sv)
16648 return varname(NULL, '$', index_targ,
16649 NULL, 0, FUV_SUBSCRIPT_NONE);
16653 /* If we got to this point it was undef on a const subscript,
16654 * so magic probably involved, e.g. $ISA[0]. Give up. */
16658 /* the SV returned by pp_multideref() was undef, if anything was */
16664 sv = PAD_SV(agg_targ);
16666 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16670 if (index_type == MDEREF_INDEX_const) {
16675 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16676 if (!he || HeVAL(he) != uninit_sv)
16680 SV * const * const svp =
16681 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16682 if (!svp || *svp != uninit_sv)
16687 ? varname(agg_gv, '%', agg_targ,
16688 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16689 : varname(agg_gv, '@', agg_targ,
16690 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16693 /* index is an var */
16695 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16697 return varname(agg_gv, '%', agg_targ,
16698 keysv, 0, FUV_SUBSCRIPT_HASH);
16701 const SSize_t index
16702 = find_array_subscript((const AV *)sv, uninit_sv);
16704 return varname(agg_gv, '@', agg_targ,
16705 NULL, index, FUV_SUBSCRIPT_ARRAY);
16709 return varname(agg_gv,
16711 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16713 NOT_REACHED; /* NOTREACHED */
16717 /* only examine RHS */
16718 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16722 o = cUNOPx(obase)->op_first;
16723 if ( o->op_type == OP_PUSHMARK
16724 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16728 if (!OpHAS_SIBLING(o)) {
16729 /* one-arg version of open is highly magical */
16731 if (o->op_type == OP_GV) { /* open FOO; */
16733 if (match && GvSV(gv) != uninit_sv)
16735 return varname(gv, '$', 0,
16736 NULL, 0, FUV_SUBSCRIPT_NONE);
16738 /* other possibilities not handled are:
16739 * open $x; or open my $x; should return '${*$x}'
16740 * open expr; should return '$'.expr ideally
16747 /* ops where $_ may be an implicit arg */
16752 if ( !(obase->op_flags & OPf_STACKED)) {
16753 if (uninit_sv == DEFSV)
16754 return newSVpvs_flags("$_", SVs_TEMP);
16755 else if (obase->op_targ
16756 && uninit_sv == PAD_SVl(obase->op_targ))
16757 return varname(NULL, '$', obase->op_targ, NULL, 0,
16758 FUV_SUBSCRIPT_NONE);
16765 match = 1; /* print etc can return undef on defined args */
16766 /* skip filehandle as it can't produce 'undef' warning */
16767 o = cUNOPx(obase)->op_first;
16768 if ((obase->op_flags & OPf_STACKED)
16770 ( o->op_type == OP_PUSHMARK
16771 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16772 o = OpSIBLING(OpSIBLING(o));
16776 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16777 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16779 /* the following ops are capable of returning PL_sv_undef even for
16780 * defined arg(s) */
16799 case OP_GETPEERNAME:
16846 case OP_SMARTMATCH:
16855 /* XXX tmp hack: these two may call an XS sub, and currently
16856 XS subs don't have a SUB entry on the context stack, so CV and
16857 pad determination goes wrong, and BAD things happen. So, just
16858 don't try to determine the value under those circumstances.
16859 Need a better fix at dome point. DAPM 11/2007 */
16865 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16866 if (gv && GvSV(gv) == uninit_sv)
16867 return newSVpvs_flags("$.", SVs_TEMP);
16872 /* def-ness of rval pos() is independent of the def-ness of its arg */
16873 if ( !(obase->op_flags & OPf_MOD))
16879 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16880 return newSVpvs_flags("${$/}", SVs_TEMP);
16885 if (!(obase->op_flags & OPf_KIDS))
16887 o = cUNOPx(obase)->op_first;
16893 /* This loop checks all the kid ops, skipping any that cannot pos-
16894 * sibly be responsible for the uninitialized value; i.e., defined
16895 * constants and ops that return nothing. If there is only one op
16896 * left that is not skipped, then we *know* it is responsible for
16897 * the uninitialized value. If there is more than one op left, we
16898 * have to look for an exact match in the while() loop below.
16899 * Note that we skip padrange, because the individual pad ops that
16900 * it replaced are still in the tree, so we work on them instead.
16903 for (kid=o; kid; kid = OpSIBLING(kid)) {
16904 const OPCODE type = kid->op_type;
16905 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16906 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16907 || (type == OP_PUSHMARK)
16908 || (type == OP_PADRANGE)
16912 if (o2) { /* more than one found */
16919 return find_uninit_var(o2, uninit_sv, match, desc_p);
16921 /* scan all args */
16923 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16935 =for apidoc report_uninit
16937 Print appropriate "Use of uninitialized variable" warning.
16943 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16945 const char *desc = NULL;
16946 SV* varname = NULL;
16949 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16951 : PL_op->op_type == OP_MULTICONCAT
16952 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16955 if (uninit_sv && PL_curpad) {
16956 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16958 sv_insert(varname, 0, 0, " ", 1);
16961 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16962 /* we've reached the end of a sort block or sub,
16963 * and the uninit value is probably what that code returned */
16966 /* PL_warn_uninit_sv is constant */
16967 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
16969 /* diag_listed_as: Use of uninitialized value%s */
16970 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16971 SVfARG(varname ? varname : &PL_sv_no),
16974 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16976 GCC_DIAG_RESTORE_STMT;
16980 * ex: set ts=8 sts=4 sw=4 et: