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.
2522 =for apidoc Amnh||SV_GMAGIC
2528 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2530 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2532 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2538 if (flags & SV_SKIP_OVERLOAD)
2540 tmpstr = AMG_CALLunary(sv, numer_amg);
2541 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2542 return SvUV(tmpstr);
2545 return PTR2UV(SvRV(sv));
2548 if (SvVALID(sv) || isREGEXP(sv)) {
2549 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2550 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2551 Regexps have no SvIVX and SvNVX fields. */
2555 const char * const ptr =
2556 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2558 = grok_number(ptr, SvCUR(sv), &value);
2560 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2561 == IS_NUMBER_IN_UV) {
2562 /* It's definitely an integer */
2563 if (!(numtype & IS_NUMBER_NEG))
2567 /* Quite wrong but no good choices. */
2568 if ((numtype & IS_NUMBER_INFINITY)) {
2569 return UV_MAX; /* So wrong. */
2570 } else if ((numtype & IS_NUMBER_NAN)) {
2571 return 0; /* So wrong. */
2575 if (ckWARN(WARN_NUMERIC))
2578 return U_V(Atof(ptr));
2582 if (SvTHINKFIRST(sv)) {
2583 if (SvREADONLY(sv) && !SvOK(sv)) {
2584 if (ckWARN(WARN_UNINITIALIZED))
2591 if (S_sv_2iuv_common(aTHX_ sv))
2595 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2596 PTR2UV(sv),SvUVX(sv)));
2597 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2601 =for apidoc sv_2nv_flags
2603 Return the num value of an SV, doing any necessary string or integer
2604 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2605 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2611 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2613 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2615 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2616 && SvTYPE(sv) != SVt_PVFM);
2617 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2618 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2619 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2620 Regexps have no SvIVX and SvNVX fields. */
2622 if (flags & SV_GMAGIC)
2626 if (SvPOKp(sv) && !SvIOKp(sv)) {
2627 ptr = SvPVX_const(sv);
2628 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2629 !grok_number(ptr, SvCUR(sv), NULL))
2635 return (NV)SvUVX(sv);
2637 return (NV)SvIVX(sv);
2642 assert(SvTYPE(sv) >= SVt_PVMG);
2643 /* This falls through to the report_uninit near the end of the
2645 } else if (SvTHINKFIRST(sv)) {
2650 if (flags & SV_SKIP_OVERLOAD)
2652 tmpstr = AMG_CALLunary(sv, numer_amg);
2653 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2654 return SvNV(tmpstr);
2657 return PTR2NV(SvRV(sv));
2659 if (SvREADONLY(sv) && !SvOK(sv)) {
2660 if (ckWARN(WARN_UNINITIALIZED))
2665 if (SvTYPE(sv) < SVt_NV) {
2666 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2667 sv_upgrade(sv, SVt_NV);
2668 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2670 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2671 STORE_LC_NUMERIC_SET_STANDARD();
2672 PerlIO_printf(Perl_debug_log,
2673 "0x%" UVxf " num(%" NVgf ")\n",
2674 PTR2UV(sv), SvNVX(sv));
2675 RESTORE_LC_NUMERIC();
2677 CLANG_DIAG_RESTORE_STMT;
2680 else if (SvTYPE(sv) < SVt_PVNV)
2681 sv_upgrade(sv, SVt_PVNV);
2686 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2687 #ifdef NV_PRESERVES_UV
2693 /* Only set the public NV OK flag if this NV preserves the IV */
2694 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2696 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2697 : (SvIVX(sv) == I_V(SvNVX(sv))))
2703 else if (SvPOKp(sv)) {
2705 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2706 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2708 #ifdef NV_PRESERVES_UV
2709 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2710 == IS_NUMBER_IN_UV) {
2711 /* It's definitely an integer */
2712 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2714 S_sv_setnv(aTHX_ sv, numtype);
2721 SvNV_set(sv, Atof(SvPVX_const(sv)));
2722 /* Only set the public NV OK flag if this NV preserves the value in
2723 the PV at least as well as an IV/UV would.
2724 Not sure how to do this 100% reliably. */
2725 /* if that shift count is out of range then Configure's test is
2726 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2728 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2729 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2730 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2731 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2732 /* Can't use strtol etc to convert this string, so don't try.
2733 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2736 /* value has been set. It may not be precise. */
2737 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2738 /* 2s complement assumption for (UV)IV_MIN */
2739 SvNOK_on(sv); /* Integer is too negative. */
2744 if (numtype & IS_NUMBER_NEG) {
2745 /* -IV_MIN is undefined, but we should never reach
2746 * this point with both IS_NUMBER_NEG and value ==
2748 assert(value != (UV)IV_MIN);
2749 SvIV_set(sv, -(IV)value);
2750 } else if (value <= (UV)IV_MAX) {
2751 SvIV_set(sv, (IV)value);
2753 SvUV_set(sv, value);
2757 if (numtype & IS_NUMBER_NOT_INT) {
2758 /* I believe that even if the original PV had decimals,
2759 they are lost beyond the limit of the FP precision.
2760 However, neither is canonical, so both only get p
2761 flags. NWC, 2000/11/25 */
2762 /* Both already have p flags, so do nothing */
2764 const NV nv = SvNVX(sv);
2765 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2766 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2767 if (SvIVX(sv) == I_V(nv)) {
2770 /* It had no "." so it must be integer. */
2774 /* between IV_MAX and NV(UV_MAX).
2775 Could be slightly > UV_MAX */
2777 if (numtype & IS_NUMBER_NOT_INT) {
2778 /* UV and NV both imprecise. */
2780 const UV nv_as_uv = U_V(nv);
2782 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2791 /* It might be more code efficient to go through the entire logic above
2792 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2793 gets complex and potentially buggy, so more programmer efficient
2794 to do it this way, by turning off the public flags: */
2796 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2797 #endif /* NV_PRESERVES_UV */
2800 if (isGV_with_GP(sv)) {
2801 glob_2number(MUTABLE_GV(sv));
2805 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2807 assert (SvTYPE(sv) >= SVt_NV);
2808 /* Typically the caller expects that sv_any is not NULL now. */
2809 /* XXX Ilya implies that this is a bug in callers that assume this
2810 and ideally should be fixed. */
2813 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2815 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2816 STORE_LC_NUMERIC_SET_STANDARD();
2817 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2818 PTR2UV(sv), SvNVX(sv));
2819 RESTORE_LC_NUMERIC();
2821 CLANG_DIAG_RESTORE_STMT;
2828 Return an SV with the numeric value of the source SV, doing any necessary
2829 reference or overload conversion. The caller is expected to have handled
2836 Perl_sv_2num(pTHX_ SV *const sv)
2838 PERL_ARGS_ASSERT_SV_2NUM;
2843 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2844 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2845 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2846 return sv_2num(tmpsv);
2848 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2851 /* int2str_table: lookup table containing string representations of all
2852 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2853 * int2str_table.arr[12*2] is "12".
2855 * We are going to read two bytes at a time, so we have to ensure that
2856 * the array is aligned to a 2 byte boundary. That's why it was made a
2857 * union with a dummy U16 member. */
2858 static const union {
2861 } int2str_table = {{
2862 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2863 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2864 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2865 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2866 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2867 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2868 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2869 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2870 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2871 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2872 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2873 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2874 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2875 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2879 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2880 * UV as a string towards the end of buf, and return pointers to start and
2883 * We assume that buf is at least TYPE_CHARS(UV) long.
2886 PERL_STATIC_INLINE char *
2887 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2889 char *ptr = buf + TYPE_CHARS(UV);
2890 char * const ebuf = ptr;
2892 U16 *word_ptr, *word_table;
2894 PERL_ARGS_ASSERT_UIV_2BUF;
2896 /* ptr has to be properly aligned, because we will cast it to U16* */
2897 assert(PTR2nat(ptr) % 2 == 0);
2898 /* we are going to read/write two bytes at a time */
2899 word_ptr = (U16*)ptr;
2900 word_table = (U16*)int2str_table.arr;
2902 if (UNLIKELY(is_uv))
2908 /* Using 0- here to silence bogus warning from MS VC */
2909 uv = (UV) (0 - (UV) iv);
2914 *--word_ptr = word_table[uv % 100];
2917 ptr = (char*)word_ptr;
2920 *--ptr = (char)uv + '0';
2922 *--word_ptr = word_table[uv];
2923 ptr = (char*)word_ptr;
2933 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2934 * infinity or a not-a-number, writes the appropriate strings to the
2935 * buffer, including a zero byte. On success returns the written length,
2936 * excluding the zero byte, on failure (not an infinity, not a nan)
2937 * returns zero, assert-fails on maxlen being too short.
2939 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2940 * shared string constants we point to, instead of generating a new
2941 * string for each instance. */
2943 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2945 assert(maxlen >= 4);
2946 if (Perl_isinf(nv)) {
2948 if (maxlen < 5) /* "-Inf\0" */
2958 else if (Perl_isnan(nv)) {
2962 /* XXX optionally output the payload mantissa bits as
2963 * "(unsigned)" (to match the nan("...") C99 function,
2964 * or maybe as "(0xhhh...)" would make more sense...
2965 * provide a format string so that the user can decide?
2966 * NOTE: would affect the maxlen and assert() logic.*/
2971 assert((s == buffer + 3) || (s == buffer + 4));
2977 =for apidoc sv_2pv_flags
2979 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2980 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2981 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2982 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2988 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2992 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2994 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2995 && SvTYPE(sv) != SVt_PVFM);
2996 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3001 if (flags & SV_SKIP_OVERLOAD)
3003 tmpstr = AMG_CALLunary(sv, string_amg);
3004 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3005 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3007 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3011 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3012 if (flags & SV_CONST_RETURN) {
3013 pv = (char *) SvPVX_const(tmpstr);
3015 pv = (flags & SV_MUTABLE_RETURN)
3016 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3019 *lp = SvCUR(tmpstr);
3021 pv = sv_2pv_flags(tmpstr, lp, flags);
3034 SV *const referent = SvRV(sv);
3038 retval = buffer = savepvn("NULLREF", len);
3039 } else if (SvTYPE(referent) == SVt_REGEXP &&
3040 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3041 amagic_is_enabled(string_amg))) {
3042 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3046 /* If the regex is UTF-8 we want the containing scalar to
3047 have an UTF-8 flag too */
3054 *lp = RX_WRAPLEN(re);
3056 return RX_WRAPPED(re);
3058 const char *const typestr = sv_reftype(referent, 0);
3059 const STRLEN typelen = strlen(typestr);
3060 UV addr = PTR2UV(referent);
3061 const char *stashname = NULL;
3062 STRLEN stashnamelen = 0; /* hush, gcc */
3063 const char *buffer_end;
3065 if (SvOBJECT(referent)) {
3066 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3069 stashname = HEK_KEY(name);
3070 stashnamelen = HEK_LEN(name);
3072 if (HEK_UTF8(name)) {
3078 stashname = "__ANON__";
3081 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3082 + 2 * sizeof(UV) + 2 /* )\0 */;
3084 len = typelen + 3 /* (0x */
3085 + 2 * sizeof(UV) + 2 /* )\0 */;
3088 Newx(buffer, len, char);
3089 buffer_end = retval = buffer + len;
3091 /* Working backwards */
3095 *--retval = PL_hexdigit[addr & 15];
3096 } while (addr >>= 4);
3102 memcpy(retval, typestr, typelen);
3106 retval -= stashnamelen;
3107 memcpy(retval, stashname, stashnamelen);
3109 /* retval may not necessarily have reached the start of the
3111 assert (retval >= buffer);
3113 len = buffer_end - retval - 1; /* -1 for that \0 */
3125 if (flags & SV_MUTABLE_RETURN)
3126 return SvPVX_mutable(sv);
3127 if (flags & SV_CONST_RETURN)
3128 return (char *)SvPVX_const(sv);
3133 /* I'm assuming that if both IV and NV are equally valid then
3134 converting the IV is going to be more efficient */
3135 const U32 isUIOK = SvIsUV(sv);
3136 /* The purpose of this union is to ensure that arr is aligned on
3137 a 2 byte boundary, because that is what uiv_2buf() requires */
3139 char arr[TYPE_CHARS(UV)];
3145 if (SvTYPE(sv) < SVt_PVIV)
3146 sv_upgrade(sv, SVt_PVIV);
3147 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3149 /* inlined from sv_setpvn */
3150 s = SvGROW_mutable(sv, len + 1);
3151 Move(ptr, s, len, char);
3156 else if (SvNOK(sv)) {
3157 if (SvTYPE(sv) < SVt_PVNV)
3158 sv_upgrade(sv, SVt_PVNV);
3159 if (SvNVX(sv) == 0.0
3160 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3161 && !Perl_isnan(SvNVX(sv))
3164 s = SvGROW_mutable(sv, 2);
3169 STRLEN size = 5; /* "-Inf\0" */
3171 s = SvGROW_mutable(sv, size);
3172 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3178 /* some Xenix systems wipe out errno here */
3187 5 + /* exponent digits */
3191 s = SvGROW_mutable(sv, size);
3192 #ifndef USE_LOCALE_NUMERIC
3193 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3199 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3200 STORE_LC_NUMERIC_SET_TO_NEEDED();
3202 local_radix = _NOT_IN_NUMERIC_STANDARD;
3203 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3204 size += SvCUR(PL_numeric_radix_sv) - 1;
3205 s = SvGROW_mutable(sv, size);
3208 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3210 /* If the radix character is UTF-8, and actually is in the
3211 * output, turn on the UTF-8 flag for the scalar */
3213 && SvUTF8(PL_numeric_radix_sv)
3214 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3219 RESTORE_LC_NUMERIC();
3222 /* We don't call SvPOK_on(), because it may come to
3223 * pass that the locale changes so that the
3224 * stringification we just did is no longer correct. We
3225 * will have to re-stringify every time it is needed */
3232 else if (isGV_with_GP(sv)) {
3233 GV *const gv = MUTABLE_GV(sv);
3234 SV *const buffer = sv_newmortal();
3236 gv_efullname3(buffer, gv, "*");
3238 assert(SvPOK(buffer));
3244 *lp = SvCUR(buffer);
3245 return SvPVX(buffer);
3250 if (flags & SV_UNDEF_RETURNS_NULL)
3252 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3254 /* Typically the caller expects that sv_any is not NULL now. */
3255 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3256 sv_upgrade(sv, SVt_PV);
3261 const STRLEN len = s - SvPVX_const(sv);
3266 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3267 PTR2UV(sv),SvPVX_const(sv)));
3268 if (flags & SV_CONST_RETURN)
3269 return (char *)SvPVX_const(sv);
3270 if (flags & SV_MUTABLE_RETURN)
3271 return SvPVX_mutable(sv);
3276 =for apidoc sv_copypv
3278 Copies a stringified representation of the source SV into the
3279 destination SV. Automatically performs any necessary C<mg_get> and
3280 coercion of numeric values into strings. Guaranteed to preserve
3281 C<UTF8> flag even from overloaded objects. Similar in nature to
3282 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3283 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3284 would lose the UTF-8'ness of the PV.
3286 =for apidoc sv_copypv_nomg
3288 Like C<sv_copypv>, but doesn't invoke get magic first.
3290 =for apidoc sv_copypv_flags
3292 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3293 has the C<SV_GMAGIC> bit set.
3299 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3304 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3306 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3307 sv_setpvn(dsv,s,len);
3315 =for apidoc sv_2pvbyte
3317 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3318 to its length. May cause the SV to be downgraded from UTF-8 as a
3321 Usually accessed via the C<SvPVbyte> macro.
3327 Perl_sv_2pvbyte_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
3329 PERL_ARGS_ASSERT_SV_2PVBYTE_FLAGS;
3331 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3333 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3334 || isGV_with_GP(sv) || SvROK(sv)) {
3335 SV *sv2 = sv_newmortal();
3336 sv_copypv_nomg(sv2,sv);
3339 sv_utf8_downgrade_nomg(sv,0);
3340 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3344 =for apidoc sv_2pvutf8
3346 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3347 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3349 Usually accessed via the C<SvPVutf8> macro.
3355 Perl_sv_2pvutf8_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
3357 PERL_ARGS_ASSERT_SV_2PVUTF8_FLAGS;
3359 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3361 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3362 || isGV_with_GP(sv) || SvROK(sv)) {
3363 SV *sv2 = sv_newmortal();
3364 sv_copypv_nomg(sv2,sv);
3367 sv_utf8_upgrade_nomg(sv);
3368 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3373 =for apidoc sv_2bool
3375 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3376 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3377 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3379 =for apidoc sv_2bool_flags
3381 This function is only used by C<sv_true()> and friends, and only if
3382 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3383 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3390 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3392 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3395 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3401 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3402 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3405 if(SvGMAGICAL(sv)) {
3407 goto restart; /* call sv_2bool */
3409 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3410 else if(!SvOK(sv)) {
3413 else if(SvPOK(sv)) {
3414 svb = SvPVXtrue(sv);
3416 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3417 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3418 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3422 goto restart; /* call sv_2bool_nomg */
3432 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3434 if (SvNOK(sv) && !SvPOK(sv))
3435 return SvNVX(sv) != 0.0;
3437 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3441 =for apidoc sv_utf8_upgrade
3443 Converts the PV of an SV to its UTF-8-encoded form.
3444 Forces the SV to string form if it is not already.
3445 Will C<mg_get> on C<sv> if appropriate.
3446 Always sets the C<SvUTF8> flag to avoid future validity checks even
3447 if the whole string is the same in UTF-8 as not.
3448 Returns the number of bytes in the converted string
3450 This is not a general purpose byte encoding to Unicode interface:
3451 use the Encode extension for that.
3453 =for apidoc sv_utf8_upgrade_nomg
3455 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3457 =for apidoc sv_utf8_upgrade_flags
3459 Converts the PV of an SV to its UTF-8-encoded form.
3460 Forces the SV to string form if it is not already.
3461 Always sets the SvUTF8 flag to avoid future validity checks even
3462 if all the bytes are invariant in UTF-8.
3463 If C<flags> has C<SV_GMAGIC> bit set,
3464 will C<mg_get> on C<sv> if appropriate, else not.
3466 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3468 Returns the number of bytes in the converted string.
3470 This is not a general purpose byte encoding to Unicode interface:
3471 use the Encode extension for that.
3473 =for apidoc sv_utf8_upgrade_flags_grow
3475 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3476 the number of unused bytes the string of C<sv> is guaranteed to have free after
3477 it upon return. This allows the caller to reserve extra space that it intends
3478 to fill, to avoid extra grows.
3480 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3481 are implemented in terms of this function.
3483 Returns the number of bytes in the converted string (not including the spares).
3487 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3488 C<NUL> isn't guaranteed due to having other routines do the work in some input
3489 cases, or if the input is already flagged as being in utf8.
3494 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3496 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3498 if (sv == &PL_sv_undef)
3500 if (!SvPOK_nog(sv)) {
3502 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3503 (void) sv_2pv_flags(sv,&len, flags);
3505 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3509 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3513 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3514 * compiled and individual nodes will remain non-utf8 even if the
3515 * stringified version of the pattern gets upgraded. Whether the
3516 * PVX of a REGEXP should be grown or we should just croak, I don't
3518 if (SvUTF8(sv) || isREGEXP(sv)) {
3519 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3524 S_sv_uncow(aTHX_ sv, 0);
3527 if (SvCUR(sv) == 0) {
3528 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3530 } else { /* Assume Latin-1/EBCDIC */
3531 /* This function could be much more efficient if we
3532 * had a FLAG in SVs to signal if there are any variant
3533 * chars in the PV. Given that there isn't such a flag
3534 * make the loop as fast as possible. */
3535 U8 * s = (U8 *) SvPVX_const(sv);
3538 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3540 /* utf8 conversion not needed because all are invariants. Mark
3541 * as UTF-8 even if no variant - saves scanning loop */
3543 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3547 /* Here, there is at least one variant (t points to the first one), so
3548 * the string should be converted to utf8. Everything from 's' to
3549 * 't - 1' will occupy only 1 byte each on output.
3551 * Note that the incoming SV may not have a trailing '\0', as certain
3552 * code in pp_formline can send us partially built SVs.
3554 * There are two main ways to convert. One is to create a new string
3555 * and go through the input starting from the beginning, appending each
3556 * converted value onto the new string as we go along. Going this
3557 * route, it's probably best to initially allocate enough space in the
3558 * string rather than possibly running out of space and having to
3559 * reallocate and then copy what we've done so far. Since everything
3560 * from 's' to 't - 1' is invariant, the destination can be initialized
3561 * with these using a fast memory copy. To be sure to allocate enough
3562 * space, one could use the worst case scenario, where every remaining
3563 * byte expands to two under UTF-8, or one could parse it and count
3564 * exactly how many do expand.
3566 * The other way is to unconditionally parse the remainder of the
3567 * string to figure out exactly how big the expanded string will be,
3568 * growing if needed. Then start at the end of the string and place
3569 * the character there at the end of the unfilled space in the expanded
3570 * one, working backwards until reaching 't'.
3572 * The problem with assuming the worst case scenario is that for very
3573 * long strings, we could allocate much more memory than actually
3574 * needed, which can create performance problems. If we have to parse
3575 * anyway, the second method is the winner as it may avoid an extra
3576 * copy. The code used to use the first method under some
3577 * circumstances, but now that there is faster variant counting on
3578 * ASCII platforms, the second method is used exclusively, eliminating
3579 * some code that no longer has to be maintained. */
3582 /* Count the total number of variants there are. We can start
3583 * just beyond the first one, which is known to be at 't' */
3584 const Size_t invariant_length = t - s;
3585 U8 * e = (U8 *) SvEND(sv);
3587 /* The length of the left overs, plus 1. */
3588 const Size_t remaining_length_p1 = e - t;
3590 /* We expand by 1 for the variant at 't' and one for each remaining
3591 * variant (we start looking at 't+1') */
3592 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3594 /* +1 = trailing NUL */
3595 Size_t need = SvCUR(sv) + expansion + extra + 1;
3598 /* Grow if needed */
3599 if (SvLEN(sv) < need) {
3600 t = invariant_length + (U8*) SvGROW(sv, need);
3601 e = t + remaining_length_p1;
3603 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3605 /* Set the NUL at the end */
3606 d = (U8 *) SvEND(sv);
3609 /* Having decremented d, it points to the position to put the
3610 * very last byte of the expanded string. Go backwards through
3611 * the string, copying and expanding as we go, stopping when we
3612 * get to the part that is invariant the rest of the way down */
3616 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3619 *d-- = UTF8_EIGHT_BIT_LO(*e);
3620 *d-- = UTF8_EIGHT_BIT_HI(*e);
3625 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3626 /* Update pos. We do it at the end rather than during
3627 * the upgrade, to avoid slowing down the common case
3628 * (upgrade without pos).
3629 * pos can be stored as either bytes or characters. Since
3630 * this was previously a byte string we can just turn off
3631 * the bytes flag. */
3632 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3634 mg->mg_flags &= ~MGf_BYTES;
3636 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3637 magic_setutf8(sv,mg); /* clear UTF8 cache */
3647 =for apidoc sv_utf8_downgrade
3649 Attempts to convert the PV of an SV from characters to bytes.
3650 If the PV contains a character that cannot fit
3651 in a byte, this conversion will fail;
3652 in this case, either returns false or, if C<fail_ok> is not
3655 This is not a general purpose Unicode to byte encoding interface:
3656 use the C<Encode> extension for that.
3658 This function process get magic on C<sv>.
3660 =for apidoc sv_utf8_downgrade_nomg
3662 Like C<sv_utf8_downgrade>, but does not process get magic on C<sv>.
3664 =for apidoc sv_utf8_downgrade_flags
3666 Like C<sv_utf8_downgrade>, but with additional C<flags>.
3667 If C<flags> has C<SV_GMAGIC> bit set, processes get magic on C<sv>.
3673 Perl_sv_utf8_downgrade_flags(pTHX_ SV *const sv, const bool fail_ok, const U32 flags)
3675 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE_FLAGS;
3677 if (SvPOKp(sv) && SvUTF8(sv)) {
3681 U32 mg_flags = flags & SV_GMAGIC;
3684 S_sv_uncow(aTHX_ sv, 0);
3686 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3688 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3689 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3690 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3691 mg_flags|SV_CONST_RETURN);
3692 mg_flags = 0; /* sv_pos_b2u does get magic */
3694 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3695 magic_setutf8(sv,mg); /* clear UTF8 cache */
3698 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3700 if (!utf8_to_bytes(s, &len)) {
3705 Perl_croak(aTHX_ "Wide character in %s",
3708 Perl_croak(aTHX_ "Wide character");
3719 =for apidoc sv_utf8_encode
3721 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3722 flag off so that it looks like octets again.
3728 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3730 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3732 if (SvREADONLY(sv)) {
3733 sv_force_normal_flags(sv, 0);
3735 (void) sv_utf8_upgrade(sv);
3740 =for apidoc sv_utf8_decode
3742 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3743 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3744 so that it looks like a character. If the PV contains only single-byte
3745 characters, the C<SvUTF8> flag stays off.
3746 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3752 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3754 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3757 const U8 *start, *c, *first_variant;
3759 /* The octets may have got themselves encoded - get them back as
3762 if (!sv_utf8_downgrade(sv, TRUE))
3765 /* it is actually just a matter of turning the utf8 flag on, but
3766 * we want to make sure everything inside is valid utf8 first.
3768 c = start = (const U8 *) SvPVX_const(sv);
3769 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3770 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3774 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3775 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3776 after this, clearing pos. Does anything on CPAN
3778 /* adjust pos to the start of a UTF8 char sequence */
3779 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3781 I32 pos = mg->mg_len;
3783 for (c = start + pos; c > start; c--) {
3784 if (UTF8_IS_START(*c))
3787 mg->mg_len = c - start;
3790 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3791 magic_setutf8(sv,mg); /* clear UTF8 cache */
3798 =for apidoc sv_setsv
3800 Copies the contents of the source SV C<ssv> into the destination SV
3801 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3802 function if the source SV needs to be reused. Does not handle 'set' magic on
3803 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3804 performs a copy-by-value, obliterating any previous content of the
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 =for apidoc sv_setsv_flags
3813 Copies the contents of the source SV C<ssv> into the destination SV
3814 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3815 function if the source SV needs to be reused. Does not handle 'set' magic.
3816 Loosely speaking, it performs a copy-by-value, obliterating any previous
3817 content of the destination.
3818 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3819 C<ssv> if appropriate, else not. If the C<flags>
3820 parameter has the C<SV_NOSTEAL> bit set then the
3821 buffers of temps will not be stolen. C<sv_setsv>
3822 and C<sv_setsv_nomg> are implemented in terms of this function.
3824 You probably want to use one of the assortment of wrappers, such as
3825 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3826 C<SvSetMagicSV_nosteal>.
3828 This is the primary function for copying scalars, and most other
3829 copy-ish functions and macros use this underneath.
3831 =for apidoc Amnh||SV_NOSTEAL
3837 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3839 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3840 HV *old_stash = NULL;
3842 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3844 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3845 const char * const name = GvNAME(sstr);
3846 const STRLEN len = GvNAMELEN(sstr);
3848 if (dtype >= SVt_PV) {
3854 SvUPGRADE(dstr, SVt_PVGV);
3855 (void)SvOK_off(dstr);
3856 isGV_with_GP_on(dstr);
3858 GvSTASH(dstr) = GvSTASH(sstr);
3860 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3861 gv_name_set(MUTABLE_GV(dstr), name, len,
3862 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3863 SvFAKE_on(dstr); /* can coerce to non-glob */
3866 if(GvGP(MUTABLE_GV(sstr))) {
3867 /* If source has method cache entry, clear it */
3869 SvREFCNT_dec(GvCV(sstr));
3870 GvCV_set(sstr, NULL);
3873 /* If source has a real method, then a method is
3876 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3882 /* If dest already had a real method, that's a change as well */
3884 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3885 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3890 /* We don't need to check the name of the destination if it was not a
3891 glob to begin with. */
3892 if(dtype == SVt_PVGV) {
3893 const char * const name = GvNAME((const GV *)dstr);
3894 const STRLEN len = GvNAMELEN(dstr);
3895 if(memEQs(name, len, "ISA")
3896 /* The stash may have been detached from the symbol table, so
3898 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3902 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3903 || (len == 1 && name[0] == ':')) {
3906 /* Set aside the old stash, so we can reset isa caches on
3908 if((old_stash = GvHV(dstr)))
3909 /* Make sure we do not lose it early. */
3910 SvREFCNT_inc_simple_void_NN(
3911 sv_2mortal((SV *)old_stash)
3916 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3919 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3920 * so temporarily protect it */
3922 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3923 gp_free(MUTABLE_GV(dstr));
3924 GvINTRO_off(dstr); /* one-shot flag */
3925 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3928 if (SvTAINTED(sstr))
3930 if (GvIMPORTED(dstr) != GVf_IMPORTED
3931 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3933 GvIMPORTED_on(dstr);
3936 if(mro_changes == 2) {
3937 if (GvAV((const GV *)sstr)) {
3939 SV * const sref = (SV *)GvAV((const GV *)dstr);
3940 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3941 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3942 AV * const ary = newAV();
3943 av_push(ary, mg->mg_obj); /* takes the refcount */
3944 mg->mg_obj = (SV *)ary;
3946 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3948 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3950 mro_isa_changed_in(GvSTASH(dstr));
3952 else if(mro_changes == 3) {
3953 HV * const stash = GvHV(dstr);
3954 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3960 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3961 if (GvIO(dstr) && dtype == SVt_PVGV) {
3962 DEBUG_o(Perl_deb(aTHX_
3963 "glob_assign_glob clearing PL_stashcache\n"));
3964 /* It's a cache. It will rebuild itself quite happily.
3965 It's a lot of effort to work out exactly which key (or keys)
3966 might be invalidated by the creation of the this file handle.
3968 hv_clear(PL_stashcache);
3974 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3976 SV * const sref = SvRV(sstr);
3978 const int intro = GvINTRO(dstr);
3981 const U32 stype = SvTYPE(sref);
3983 PERL_ARGS_ASSERT_GV_SETREF;
3986 GvINTRO_off(dstr); /* one-shot flag */
3987 GvLINE(dstr) = CopLINE(PL_curcop);
3988 GvEGV(dstr) = MUTABLE_GV(dstr);
3993 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3994 import_flag = GVf_IMPORTED_CV;
3997 location = (SV **) &GvHV(dstr);
3998 import_flag = GVf_IMPORTED_HV;
4001 location = (SV **) &GvAV(dstr);
4002 import_flag = GVf_IMPORTED_AV;
4005 location = (SV **) &GvIOp(dstr);
4008 location = (SV **) &GvFORM(dstr);
4011 location = &GvSV(dstr);
4012 import_flag = GVf_IMPORTED_SV;
4015 if (stype == SVt_PVCV) {
4016 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4017 if (GvCVGEN(dstr)) {
4018 SvREFCNT_dec(GvCV(dstr));
4019 GvCV_set(dstr, NULL);
4020 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4023 /* SAVEt_GVSLOT takes more room on the savestack and has more
4024 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4025 leave_scope needs access to the GV so it can reset method
4026 caches. We must use SAVEt_GVSLOT whenever the type is
4027 SVt_PVCV, even if the stash is anonymous, as the stash may
4028 gain a name somehow before leave_scope. */
4029 if (stype == SVt_PVCV) {
4030 /* There is no save_pushptrptrptr. Creating it for this
4031 one call site would be overkill. So inline the ss add
4035 SS_ADD_PTR(location);
4036 SS_ADD_PTR(SvREFCNT_inc(*location));
4037 SS_ADD_UV(SAVEt_GVSLOT);
4040 else SAVEGENERICSV(*location);
4043 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4044 CV* const cv = MUTABLE_CV(*location);
4046 if (!GvCVGEN((const GV *)dstr) &&
4047 (CvROOT(cv) || CvXSUB(cv)) &&
4048 /* redundant check that avoids creating the extra SV
4049 most of the time: */
4050 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4052 SV * const new_const_sv =
4053 CvCONST((const CV *)sref)
4054 ? cv_const_sv((const CV *)sref)
4056 HV * const stash = GvSTASH((const GV *)dstr);
4057 report_redefined_cv(
4060 ? Perl_newSVpvf(aTHX_
4061 "%" HEKf "::%" HEKf,
4062 HEKfARG(HvNAME_HEK(stash)),
4063 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4064 : Perl_newSVpvf(aTHX_
4066 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4069 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4073 cv_ckproto_len_flags(cv, (const GV *)dstr,
4074 SvPOK(sref) ? CvPROTO(sref) : NULL,
4075 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4076 SvPOK(sref) ? SvUTF8(sref) : 0);
4078 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4079 GvASSUMECV_on(dstr);
4080 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4081 if (intro && GvREFCNT(dstr) > 1) {
4082 /* temporary remove extra savestack's ref */
4084 gv_method_changed(dstr);
4087 else gv_method_changed(dstr);
4090 *location = SvREFCNT_inc_simple_NN(sref);
4091 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4092 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4093 GvFLAGS(dstr) |= import_flag;
4096 if (stype == SVt_PVHV) {
4097 const char * const name = GvNAME((GV*)dstr);
4098 const STRLEN len = GvNAMELEN(dstr);
4101 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4102 || (len == 1 && name[0] == ':')
4104 && (!dref || HvENAME_get(dref))
4107 (HV *)sref, (HV *)dref,
4113 stype == SVt_PVAV && sref != dref
4114 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4115 /* The stash may have been detached from the symbol table, so
4116 check its name before doing anything. */
4117 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4120 MAGIC * const omg = dref && SvSMAGICAL(dref)
4121 ? mg_find(dref, PERL_MAGIC_isa)
4123 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4124 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4125 AV * const ary = newAV();
4126 av_push(ary, mg->mg_obj); /* takes the refcount */
4127 mg->mg_obj = (SV *)ary;
4130 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4131 SV **svp = AvARRAY((AV *)omg->mg_obj);
4132 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4136 SvREFCNT_inc_simple_NN(*svp++)
4142 SvREFCNT_inc_simple_NN(omg->mg_obj)
4146 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4152 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4154 for (i = 0; i <= AvFILL(sref); ++i) {
4155 SV **elem = av_fetch ((AV*)sref, i, 0);
4158 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4162 mg = mg_find(sref, PERL_MAGIC_isa);
4164 /* Since the *ISA assignment could have affected more than
4165 one stash, don't call mro_isa_changed_in directly, but let
4166 magic_clearisa do it for us, as it already has the logic for
4167 dealing with globs vs arrays of globs. */
4169 Perl_magic_clearisa(aTHX_ NULL, mg);
4171 else if (stype == SVt_PVIO) {
4172 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4173 /* It's a cache. It will rebuild itself quite happily.
4174 It's a lot of effort to work out exactly which key (or keys)
4175 might be invalidated by the creation of the this file handle.
4177 hv_clear(PL_stashcache);
4181 if (!intro) SvREFCNT_dec(dref);
4182 if (SvTAINTED(sstr))
4190 #ifdef PERL_DEBUG_READONLY_COW
4191 # include <sys/mman.h>
4193 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4194 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4198 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4200 struct perl_memory_debug_header * const header =
4201 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4202 const MEM_SIZE len = header->size;
4203 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4204 # ifdef PERL_TRACK_MEMPOOL
4205 if (!header->readonly) header->readonly = 1;
4207 if (mprotect(header, len, PROT_READ))
4208 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4209 header, len, errno);
4213 S_sv_buf_to_rw(pTHX_ SV *sv)
4215 struct perl_memory_debug_header * const header =
4216 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4217 const MEM_SIZE len = header->size;
4218 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4219 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4220 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4221 header, len, errno);
4222 # ifdef PERL_TRACK_MEMPOOL
4223 header->readonly = 0;
4228 # define sv_buf_to_ro(sv) NOOP
4229 # define sv_buf_to_rw(sv) NOOP
4233 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4238 unsigned int both_type;
4240 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4242 if (UNLIKELY( sstr == dstr ))
4245 if (UNLIKELY( !sstr ))
4246 sstr = &PL_sv_undef;
4248 stype = SvTYPE(sstr);
4249 dtype = SvTYPE(dstr);
4250 both_type = (stype | dtype);
4252 /* with these values, we can check that both SVs are NULL/IV (and not
4253 * freed) just by testing the or'ed types */
4254 STATIC_ASSERT_STMT(SVt_NULL == 0);
4255 STATIC_ASSERT_STMT(SVt_IV == 1);
4256 if (both_type <= 1) {
4257 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4263 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4264 if (SvREADONLY(dstr))
4265 Perl_croak_no_modify();
4267 if (SvWEAKREF(dstr))
4268 sv_unref_flags(dstr, 0);
4270 old_rv = SvRV(dstr);
4273 assert(!SvGMAGICAL(sstr));
4274 assert(!SvGMAGICAL(dstr));
4276 sflags = SvFLAGS(sstr);
4277 if (sflags & (SVf_IOK|SVf_ROK)) {
4278 SET_SVANY_FOR_BODYLESS_IV(dstr);
4279 new_dflags = SVt_IV;
4281 if (sflags & SVf_ROK) {
4282 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4283 new_dflags |= SVf_ROK;
4286 /* both src and dst are <= SVt_IV, so sv_any points to the
4287 * head; so access the head directly
4289 assert( &(sstr->sv_u.svu_iv)
4290 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4291 assert( &(dstr->sv_u.svu_iv)
4292 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4293 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4294 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4298 new_dflags = dtype; /* turn off everything except the type */
4300 SvFLAGS(dstr) = new_dflags;
4301 SvREFCNT_dec(old_rv);
4306 if (UNLIKELY(both_type == SVTYPEMASK)) {
4307 if (SvIS_FREED(dstr)) {
4308 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4309 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4311 if (SvIS_FREED(sstr)) {
4312 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4313 (void*)sstr, (void*)dstr);
4319 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4320 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4322 /* There's a lot of redundancy below but we're going for speed here */
4327 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4328 (void)SvOK_off(dstr);
4336 /* For performance, we inline promoting to type SVt_IV. */
4337 /* We're starting from SVt_NULL, so provided that define is
4338 * actual 0, we don't have to unset any SV type flags
4339 * to promote to SVt_IV. */
4340 STATIC_ASSERT_STMT(SVt_NULL == 0);
4341 SET_SVANY_FOR_BODYLESS_IV(dstr);
4342 SvFLAGS(dstr) |= SVt_IV;
4346 sv_upgrade(dstr, SVt_PVIV);
4350 goto end_of_first_switch;
4352 (void)SvIOK_only(dstr);
4353 SvIV_set(dstr, SvIVX(sstr));
4356 /* SvTAINTED can only be true if the SV has taint magic, which in
4357 turn means that the SV type is PVMG (or greater). This is the
4358 case statement for SVt_IV, so this cannot be true (whatever gcov
4360 assert(!SvTAINTED(sstr));
4365 if (dtype < SVt_PV && dtype != SVt_IV)
4366 sv_upgrade(dstr, SVt_IV);
4370 if (LIKELY( SvNOK(sstr) )) {
4374 sv_upgrade(dstr, SVt_NV);
4378 sv_upgrade(dstr, SVt_PVNV);
4382 goto end_of_first_switch;
4384 SvNV_set(dstr, SvNVX(sstr));
4385 (void)SvNOK_only(dstr);
4386 /* SvTAINTED can only be true if the SV has taint magic, which in
4387 turn means that the SV type is PVMG (or greater). This is the
4388 case statement for SVt_NV, so this cannot be true (whatever gcov
4390 assert(!SvTAINTED(sstr));
4397 sv_upgrade(dstr, SVt_PV);
4400 if (dtype < SVt_PVIV)
4401 sv_upgrade(dstr, SVt_PVIV);
4404 if (dtype < SVt_PVNV)
4405 sv_upgrade(dstr, SVt_PVNV);
4409 invlist_clone(sstr, dstr);
4413 const char * const type = sv_reftype(sstr,0);
4415 /* diag_listed_as: Bizarre copy of %s */
4416 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4418 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4420 NOT_REACHED; /* NOTREACHED */
4424 if (dtype < SVt_REGEXP)
4425 sv_upgrade(dstr, SVt_REGEXP);
4431 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4433 if (SvTYPE(sstr) != stype)
4434 stype = SvTYPE(sstr);
4436 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4437 glob_assign_glob(dstr, sstr, dtype);
4440 if (stype == SVt_PVLV)
4442 if (isREGEXP(sstr)) goto upgregexp;
4443 SvUPGRADE(dstr, SVt_PVNV);
4446 SvUPGRADE(dstr, (svtype)stype);
4448 end_of_first_switch:
4450 /* dstr may have been upgraded. */
4451 dtype = SvTYPE(dstr);
4452 sflags = SvFLAGS(sstr);
4454 if (UNLIKELY( dtype == SVt_PVCV )) {
4455 /* Assigning to a subroutine sets the prototype. */
4458 const char *const ptr = SvPV_const(sstr, len);
4460 SvGROW(dstr, len + 1);
4461 Copy(ptr, SvPVX(dstr), len + 1, char);
4462 SvCUR_set(dstr, len);
4464 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4465 CvAUTOLOAD_off(dstr);
4470 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4471 || dtype == SVt_PVFM))
4473 const char * const type = sv_reftype(dstr,0);
4475 /* diag_listed_as: Cannot copy to %s */
4476 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4478 Perl_croak(aTHX_ "Cannot copy to %s", type);
4479 } else if (sflags & SVf_ROK) {
4480 if (isGV_with_GP(dstr)
4481 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4484 if (GvIMPORTED(dstr) != GVf_IMPORTED
4485 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4487 GvIMPORTED_on(dstr);
4492 glob_assign_glob(dstr, sstr, dtype);
4496 if (dtype >= SVt_PV) {
4497 if (isGV_with_GP(dstr)) {
4498 gv_setref(dstr, sstr);
4501 if (SvPVX_const(dstr)) {
4507 (void)SvOK_off(dstr);
4508 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4509 SvFLAGS(dstr) |= sflags & SVf_ROK;
4510 assert(!(sflags & SVp_NOK));
4511 assert(!(sflags & SVp_IOK));
4512 assert(!(sflags & SVf_NOK));
4513 assert(!(sflags & SVf_IOK));
4515 else if (isGV_with_GP(dstr)) {
4516 if (!(sflags & SVf_OK)) {
4517 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4518 "Undefined value assigned to typeglob");
4521 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4522 if (dstr != (const SV *)gv) {
4523 const char * const name = GvNAME((const GV *)dstr);
4524 const STRLEN len = GvNAMELEN(dstr);
4525 HV *old_stash = NULL;
4526 bool reset_isa = FALSE;
4527 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4528 || (len == 1 && name[0] == ':')) {
4529 /* Set aside the old stash, so we can reset isa caches
4530 on its subclasses. */
4531 if((old_stash = GvHV(dstr))) {
4532 /* Make sure we do not lose it early. */
4533 SvREFCNT_inc_simple_void_NN(
4534 sv_2mortal((SV *)old_stash)
4541 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4542 gp_free(MUTABLE_GV(dstr));
4544 GvGP_set(dstr, gp_ref(GvGP(gv)));
4547 HV * const stash = GvHV(dstr);
4549 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4559 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4560 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4561 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4563 else if (sflags & SVp_POK) {
4564 const STRLEN cur = SvCUR(sstr);
4565 const STRLEN len = SvLEN(sstr);
4568 * We have three basic ways to copy the string:
4574 * Which we choose is based on various factors. The following
4575 * things are listed in order of speed, fastest to slowest:
4577 * - Copying a short string
4578 * - Copy-on-write bookkeeping
4580 * - Copying a long string
4582 * We swipe the string (steal the string buffer) if the SV on the
4583 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4584 * big win on long strings. It should be a win on short strings if
4585 * SvPVX_const(dstr) has to be allocated. If not, it should not
4586 * slow things down, as SvPVX_const(sstr) would have been freed
4589 * We also steal the buffer from a PADTMP (operator target) if it
4590 * is ‘long enough’. For short strings, a swipe does not help
4591 * here, as it causes more malloc calls the next time the target
4592 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4593 * be allocated it is still not worth swiping PADTMPs for short
4594 * strings, as the savings here are small.
4596 * If swiping is not an option, then we see whether it is
4597 * worth using copy-on-write. If the lhs already has a buf-
4598 * fer big enough and the string is short, we skip it and fall back
4599 * to method 3, since memcpy is faster for short strings than the
4600 * later bookkeeping overhead that copy-on-write entails.
4602 * If the rhs is not a copy-on-write string yet, then we also
4603 * consider whether the buffer is too large relative to the string
4604 * it holds. Some operations such as readline allocate a large
4605 * buffer in the expectation of reusing it. But turning such into
4606 * a COW buffer is counter-productive because it increases memory
4607 * usage by making readline allocate a new large buffer the sec-
4608 * ond time round. So, if the buffer is too large, again, we use
4611 * Finally, if there is no buffer on the left, or the buffer is too
4612 * small, then we use copy-on-write and make both SVs share the
4617 /* Whichever path we take through the next code, we want this true,
4618 and doing it now facilitates the COW check. */
4619 (void)SvPOK_only(dstr);
4623 /* slated for free anyway (and not COW)? */
4624 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4625 /* or a swipable TARG */
4627 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4629 /* whose buffer is worth stealing */
4630 && CHECK_COWBUF_THRESHOLD(cur,len)
4633 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4634 (!(flags & SV_NOSTEAL)) &&
4635 /* and we're allowed to steal temps */
4636 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4637 len) /* and really is a string */
4638 { /* Passes the swipe test. */
4639 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4641 SvPV_set(dstr, SvPVX_mutable(sstr));
4642 SvLEN_set(dstr, SvLEN(sstr));
4643 SvCUR_set(dstr, SvCUR(sstr));
4646 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4647 SvPV_set(sstr, NULL);
4652 else if (flags & SV_COW_SHARED_HASH_KEYS
4654 #ifdef PERL_COPY_ON_WRITE
4657 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4658 /* If this is a regular (non-hek) COW, only so
4659 many COW "copies" are possible. */
4660 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4661 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4662 && !(SvFLAGS(dstr) & SVf_BREAK)
4663 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4664 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4668 && !(SvFLAGS(dstr) & SVf_BREAK)
4671 /* Either it's a shared hash key, or it's suitable for
4675 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4681 if (!(sflags & SVf_IsCOW)) {
4683 CowREFCNT(sstr) = 0;
4686 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4692 if (sflags & SVf_IsCOW) {
4696 SvPV_set(dstr, SvPVX_mutable(sstr));
4701 /* SvIsCOW_shared_hash */
4702 DEBUG_C(PerlIO_printf(Perl_debug_log,
4703 "Copy on write: Sharing hash\n"));
4705 assert (SvTYPE(dstr) >= SVt_PV);
4707 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4709 SvLEN_set(dstr, len);
4710 SvCUR_set(dstr, cur);
4713 /* Failed the swipe test, and we cannot do copy-on-write either.
4714 Have to copy the string. */
4715 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4716 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4717 SvCUR_set(dstr, cur);
4718 *SvEND(dstr) = '\0';
4720 if (sflags & SVp_NOK) {
4721 SvNV_set(dstr, SvNVX(sstr));
4723 if (sflags & SVp_IOK) {
4724 SvIV_set(dstr, SvIVX(sstr));
4725 if (sflags & SVf_IVisUV)
4728 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4730 const MAGIC * const smg = SvVSTRING_mg(sstr);
4732 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4733 smg->mg_ptr, smg->mg_len);
4734 SvRMAGICAL_on(dstr);
4738 else if (sflags & (SVp_IOK|SVp_NOK)) {
4739 (void)SvOK_off(dstr);
4740 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4741 if (sflags & SVp_IOK) {
4742 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4743 SvIV_set(dstr, SvIVX(sstr));
4745 if (sflags & SVp_NOK) {
4746 SvNV_set(dstr, SvNVX(sstr));
4750 if (isGV_with_GP(sstr)) {
4751 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4754 (void)SvOK_off(dstr);
4756 if (SvTAINTED(sstr))
4762 =for apidoc sv_set_undef
4764 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4765 Doesn't handle set magic.
4767 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4768 buffer, unlike C<undef $sv>.
4770 Introduced in perl 5.25.12.
4776 Perl_sv_set_undef(pTHX_ SV *sv)
4778 U32 type = SvTYPE(sv);
4780 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4782 /* shortcut, NULL, IV, RV */
4784 if (type <= SVt_IV) {
4785 assert(!SvGMAGICAL(sv));
4786 if (SvREADONLY(sv)) {
4787 /* does undeffing PL_sv_undef count as modifying a read-only
4788 * variable? Some XS code does this */
4789 if (sv == &PL_sv_undef)
4791 Perl_croak_no_modify();
4796 sv_unref_flags(sv, 0);
4799 SvFLAGS(sv) = type; /* quickly turn off all flags */
4800 SvREFCNT_dec_NN(rv);
4804 SvFLAGS(sv) = type; /* quickly turn off all flags */
4809 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4812 SV_CHECK_THINKFIRST_COW_DROP(sv);
4814 if (isGV_with_GP(sv))
4815 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4816 "Undefined value assigned to typeglob");
4824 =for apidoc sv_setsv_mg
4826 Like C<sv_setsv>, but also handles 'set' magic.
4832 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4834 PERL_ARGS_ASSERT_SV_SETSV_MG;
4836 sv_setsv(dstr,sstr);
4841 # define SVt_COW SVt_PV
4843 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4845 STRLEN cur = SvCUR(sstr);
4846 STRLEN len = SvLEN(sstr);
4848 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4849 const bool already = cBOOL(SvIsCOW(sstr));
4852 PERL_ARGS_ASSERT_SV_SETSV_COW;
4855 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4856 (void*)sstr, (void*)dstr);
4863 if (SvTHINKFIRST(dstr))
4864 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4865 else if (SvPVX_const(dstr))
4866 Safefree(SvPVX_mutable(dstr));
4870 SvUPGRADE(dstr, SVt_COW);
4872 assert (SvPOK(sstr));
4873 assert (SvPOKp(sstr));
4875 if (SvIsCOW(sstr)) {
4877 if (SvLEN(sstr) == 0) {
4878 /* source is a COW shared hash key. */
4879 DEBUG_C(PerlIO_printf(Perl_debug_log,
4880 "Fast copy on write: Sharing hash\n"));
4881 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4884 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4885 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4887 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4888 SvUPGRADE(sstr, SVt_COW);
4890 DEBUG_C(PerlIO_printf(Perl_debug_log,
4891 "Fast copy on write: Converting sstr to COW\n"));
4892 CowREFCNT(sstr) = 0;
4894 # ifdef PERL_DEBUG_READONLY_COW
4895 if (already) sv_buf_to_rw(sstr);
4898 new_pv = SvPVX_mutable(sstr);
4902 SvPV_set(dstr, new_pv);
4903 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4906 SvLEN_set(dstr, len);
4907 SvCUR_set(dstr, cur);
4917 =for apidoc sv_setpv_bufsize
4919 Sets the SV to be a string of cur bytes length, with at least
4920 len bytes available. Ensures that there is a null byte at SvEND.
4921 Returns a char * pointer to the SvPV buffer.
4927 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4931 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4933 SV_CHECK_THINKFIRST_COW_DROP(sv);
4934 SvUPGRADE(sv, SVt_PV);
4935 pv = SvGROW(sv, len + 1);
4938 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4941 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4946 =for apidoc sv_setpvn
4948 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4949 The C<len> parameter indicates the number of
4950 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4951 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4957 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4961 PERL_ARGS_ASSERT_SV_SETPVN;
4963 SV_CHECK_THINKFIRST_COW_DROP(sv);
4964 if (isGV_with_GP(sv))
4965 Perl_croak_no_modify();
4971 /* len is STRLEN which is unsigned, need to copy to signed */
4974 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4977 SvUPGRADE(sv, SVt_PV);
4979 dptr = SvGROW(sv, len + 1);
4980 Move(ptr,dptr,len,char);
4983 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4985 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4989 =for apidoc sv_setpvn_mg
4991 Like C<sv_setpvn>, but also handles 'set' magic.
4997 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4999 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5001 sv_setpvn(sv,ptr,len);
5006 =for apidoc sv_setpv
5008 Copies a string into an SV. The string must be terminated with a C<NUL>
5009 character, and not contain embeded C<NUL>'s.
5010 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5016 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5020 PERL_ARGS_ASSERT_SV_SETPV;
5022 SV_CHECK_THINKFIRST_COW_DROP(sv);
5028 SvUPGRADE(sv, SVt_PV);
5030 SvGROW(sv, len + 1);
5031 Move(ptr,SvPVX(sv),len+1,char);
5033 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5035 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5039 =for apidoc sv_setpv_mg
5041 Like C<sv_setpv>, but also handles 'set' magic.
5047 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5049 PERL_ARGS_ASSERT_SV_SETPV_MG;
5056 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5058 PERL_ARGS_ASSERT_SV_SETHEK;
5064 if (HEK_LEN(hek) == HEf_SVKEY) {
5065 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5068 const int flags = HEK_FLAGS(hek);
5069 if (flags & HVhek_WASUTF8) {
5070 STRLEN utf8_len = HEK_LEN(hek);
5071 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5072 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5075 } else if (flags & HVhek_UNSHARED) {
5076 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5079 else SvUTF8_off(sv);
5083 SV_CHECK_THINKFIRST_COW_DROP(sv);
5084 SvUPGRADE(sv, SVt_PV);
5086 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5087 SvCUR_set(sv, HEK_LEN(hek));
5093 else SvUTF8_off(sv);
5101 =for apidoc sv_usepvn_flags
5103 Tells an SV to use C<ptr> to find its string value. Normally the
5104 string is stored inside the SV, but sv_usepvn allows the SV to use an
5105 outside string. C<ptr> should point to memory that was allocated
5106 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5107 the start of a C<Newx>-ed block of memory, and not a pointer to the
5108 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5109 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5110 string length, C<len>, must be supplied. By default this function
5111 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5112 so that pointer should not be freed or used by the programmer after
5113 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5114 that pointer (e.g. ptr + 1) be used.
5116 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5117 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5119 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5120 C<len>, and already meets the requirements for storing in C<SvPVX>).
5122 =for apidoc Amnh||SV_SMAGIC
5123 =for apidoc Amnh||SV_HAS_TRAILING_NUL
5129 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5133 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5135 SV_CHECK_THINKFIRST_COW_DROP(sv);
5136 SvUPGRADE(sv, SVt_PV);
5139 if (flags & SV_SMAGIC)
5143 if (SvPVX_const(sv))
5147 if (flags & SV_HAS_TRAILING_NUL)
5148 assert(ptr[len] == '\0');
5151 allocate = (flags & SV_HAS_TRAILING_NUL)
5153 #ifdef Perl_safesysmalloc_size
5156 PERL_STRLEN_ROUNDUP(len + 1);
5158 if (flags & SV_HAS_TRAILING_NUL) {
5159 /* It's long enough - do nothing.
5160 Specifically Perl_newCONSTSUB is relying on this. */
5163 /* Force a move to shake out bugs in callers. */
5164 char *new_ptr = (char*)safemalloc(allocate);
5165 Copy(ptr, new_ptr, len, char);
5166 PoisonFree(ptr,len,char);
5170 ptr = (char*) saferealloc (ptr, allocate);
5173 #ifdef Perl_safesysmalloc_size
5174 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5176 SvLEN_set(sv, allocate);
5180 if (!(flags & SV_HAS_TRAILING_NUL)) {
5183 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5185 if (flags & SV_SMAGIC)
5191 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5193 assert(SvIsCOW(sv));
5196 const char * const pvx = SvPVX_const(sv);
5197 const STRLEN len = SvLEN(sv);
5198 const STRLEN cur = SvCUR(sv);
5202 PerlIO_printf(Perl_debug_log,
5203 "Copy on write: Force normal %ld\n",
5209 # ifdef PERL_COPY_ON_WRITE
5211 /* Must do this first, since the CowREFCNT uses SvPVX and
5212 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5213 the only owner left of the buffer. */
5214 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5216 U8 cowrefcnt = CowREFCNT(sv);
5217 if(cowrefcnt != 0) {
5219 CowREFCNT(sv) = cowrefcnt;
5224 /* Else we are the only owner of the buffer. */
5229 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5234 if (flags & SV_COW_DROP_PV) {
5235 /* OK, so we don't need to copy our buffer. */
5238 SvGROW(sv, cur + 1);
5239 Move(pvx,SvPVX(sv),cur,char);
5244 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5252 const char * const pvx = SvPVX_const(sv);
5253 const STRLEN len = SvCUR(sv);
5257 if (flags & SV_COW_DROP_PV) {
5258 /* OK, so we don't need to copy our buffer. */
5261 SvGROW(sv, len + 1);
5262 Move(pvx,SvPVX(sv),len,char);
5265 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5272 =for apidoc sv_force_normal_flags
5274 Undo various types of fakery on an SV, where fakery means
5275 "more than" a string: if the PV is a shared string, make
5276 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5277 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5278 we do the copy, and is also used locally; if this is a
5279 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5280 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5281 C<SvPOK_off> rather than making a copy. (Used where this
5282 scalar is about to be set to some other value.) In addition,
5283 the C<flags> parameter gets passed to C<sv_unref_flags()>
5284 when unreffing. C<sv_force_normal> calls this function
5285 with flags set to 0.
5287 This function is expected to be used to signal to perl that this SV is
5288 about to be written to, and any extra book-keeping needs to be taken care
5289 of. Hence, it croaks on read-only values.
5291 =for apidoc Amnh||SV_COW_DROP_PV
5297 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5299 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5302 Perl_croak_no_modify();
5303 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5304 S_sv_uncow(aTHX_ sv, flags);
5306 sv_unref_flags(sv, flags);
5307 else if (SvFAKE(sv) && isGV_with_GP(sv))
5308 sv_unglob(sv, flags);
5309 else if (SvFAKE(sv) && isREGEXP(sv)) {
5310 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5311 to sv_unglob. We only need it here, so inline it. */
5312 const bool islv = SvTYPE(sv) == SVt_PVLV;
5313 const svtype new_type =
5314 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5315 SV *const temp = newSV_type(new_type);
5316 regexp *old_rx_body;
5318 if (new_type == SVt_PVMG) {
5319 SvMAGIC_set(temp, SvMAGIC(sv));
5320 SvMAGIC_set(sv, NULL);
5321 SvSTASH_set(temp, SvSTASH(sv));
5322 SvSTASH_set(sv, NULL);
5325 SvCUR_set(temp, SvCUR(sv));
5326 /* Remember that SvPVX is in the head, not the body. */
5327 assert(ReANY((REGEXP *)sv)->mother_re);
5330 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5331 * whose xpvlenu_rx field points to the regex body */
5332 XPV *xpv = (XPV*)(SvANY(sv));
5333 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5334 xpv->xpv_len_u.xpvlenu_rx = NULL;
5337 old_rx_body = ReANY((REGEXP *)sv);
5339 /* Their buffer is already owned by someone else. */
5340 if (flags & SV_COW_DROP_PV) {
5341 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5342 zeroed body. For SVt_PVLV, we zeroed it above (len field
5343 a union with xpvlenu_rx) */
5344 assert(!SvLEN(islv ? sv : temp));
5345 sv->sv_u.svu_pv = 0;
5348 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5349 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5353 /* Now swap the rest of the bodies. */
5357 SvFLAGS(sv) &= ~SVTYPEMASK;
5358 SvFLAGS(sv) |= new_type;
5359 SvANY(sv) = SvANY(temp);
5362 SvFLAGS(temp) &= ~(SVTYPEMASK);
5363 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5364 SvANY(temp) = old_rx_body;
5366 SvREFCNT_dec_NN(temp);
5368 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5374 Efficient removal of characters from the beginning of the string buffer.
5375 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5376 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5377 character of the adjusted string. Uses the C<OOK> hack. On return, only
5378 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5380 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5381 refer to the same chunk of data.
5383 The unfortunate similarity of this function's name to that of Perl's C<chop>
5384 operator is strictly coincidental. This function works from the left;
5385 C<chop> works from the right.
5391 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5402 PERL_ARGS_ASSERT_SV_CHOP;
5404 if (!ptr || !SvPOKp(sv))
5406 delta = ptr - SvPVX_const(sv);
5408 /* Nothing to do. */
5411 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5412 if (delta > max_delta)
5413 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5414 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5415 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5416 SV_CHECK_THINKFIRST(sv);
5417 SvPOK_only_UTF8(sv);
5420 if (!SvLEN(sv)) { /* make copy of shared string */
5421 const char *pvx = SvPVX_const(sv);
5422 const STRLEN len = SvCUR(sv);
5423 SvGROW(sv, len + 1);
5424 Move(pvx,SvPVX(sv),len,char);
5430 SvOOK_offset(sv, old_delta);
5432 SvLEN_set(sv, SvLEN(sv) - delta);
5433 SvCUR_set(sv, SvCUR(sv) - delta);
5434 SvPV_set(sv, SvPVX(sv) + delta);
5436 p = (U8 *)SvPVX_const(sv);
5439 /* how many bytes were evacuated? we will fill them with sentinel
5440 bytes, except for the part holding the new offset of course. */
5443 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5445 assert(evacn <= delta + old_delta);
5449 /* This sets 'delta' to the accumulated value of all deltas so far */
5453 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5454 * the string; otherwise store a 0 byte there and store 'delta' just prior
5455 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5456 * portion of the chopped part of the string */
5457 if (delta < 0x100) {
5461 p -= sizeof(STRLEN);
5462 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5466 /* Fill the preceding buffer with sentinals to verify that no-one is
5476 =for apidoc sv_catpvn
5478 Concatenates the string onto the end of the string which is in the SV.
5479 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5480 status set, then the bytes appended should be valid UTF-8.
5481 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5483 =for apidoc sv_catpvn_flags
5485 Concatenates the string onto the end of the string which is in the SV. The
5486 C<len> indicates number of bytes to copy.
5488 By default, the string appended is assumed to be valid UTF-8 if the SV has
5489 the UTF-8 status set, and a string of bytes otherwise. One can force the
5490 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5491 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5492 string appended will be upgraded to UTF-8 if necessary.
5494 If C<flags> has the C<SV_SMAGIC> bit set, will
5495 C<mg_set> on C<dsv> afterwards if appropriate.
5496 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5497 in terms of this function.
5503 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5506 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5508 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5509 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5511 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5512 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5513 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5516 else SvGROW(dsv, dlen + slen + 3);
5518 sstr = SvPVX_const(dsv);
5519 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5520 SvCUR_set(dsv, SvCUR(dsv) + slen);
5523 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5524 const char * const send = sstr + slen;
5527 /* Something this code does not account for, which I think is
5528 impossible; it would require the same pv to be treated as
5529 bytes *and* utf8, which would indicate a bug elsewhere. */
5530 assert(sstr != dstr);
5532 SvGROW(dsv, dlen + slen * 2 + 3);
5533 d = (U8 *)SvPVX(dsv) + dlen;
5535 while (sstr < send) {
5536 append_utf8_from_native_byte(*sstr, &d);
5539 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5542 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5544 if (flags & SV_SMAGIC)
5549 =for apidoc sv_catsv
5551 Concatenates the string from SV C<ssv> onto the end of the string in SV
5552 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5553 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5554 and C<L</sv_catsv_nomg>>.
5556 =for apidoc sv_catsv_flags
5558 Concatenates the string from SV C<ssv> onto the end of the string in SV
5559 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5560 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5561 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5562 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5563 and C<sv_catsv_mg> are implemented in terms of this function.
5568 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5570 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5574 const char *spv = SvPV_flags_const(ssv, slen, flags);
5575 if (flags & SV_GMAGIC)
5577 sv_catpvn_flags(dsv, spv, slen,
5578 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5579 if (flags & SV_SMAGIC)
5585 =for apidoc sv_catpv
5587 Concatenates the C<NUL>-terminated string onto the end of the string which is
5589 If the SV has the UTF-8 status set, then the bytes appended should be
5590 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5596 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5602 PERL_ARGS_ASSERT_SV_CATPV;
5606 junk = SvPV_force(sv, tlen);
5608 SvGROW(sv, tlen + len + 1);
5610 ptr = SvPVX_const(sv);
5611 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5612 SvCUR_set(sv, SvCUR(sv) + len);
5613 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5618 =for apidoc sv_catpv_flags
5620 Concatenates the C<NUL>-terminated string onto the end of the string which is
5622 If the SV has the UTF-8 status set, then the bytes appended should
5623 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5624 on the modified SV if appropriate.
5630 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5632 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5633 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5637 =for apidoc sv_catpv_mg
5639 Like C<sv_catpv>, but also handles 'set' magic.
5645 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5647 PERL_ARGS_ASSERT_SV_CATPV_MG;
5656 Creates a new SV. A non-zero C<len> parameter indicates the number of
5657 bytes of preallocated string space the SV should have. An extra byte for a
5658 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5659 space is allocated.) The reference count for the new SV is set to 1.
5661 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5662 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5663 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5664 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5665 modules supporting older perls.
5671 Perl_newSV(pTHX_ const STRLEN len)
5677 sv_grow(sv, len + 1);
5682 =for apidoc sv_magicext
5684 Adds magic to an SV, upgrading it if necessary. Applies the
5685 supplied C<vtable> and returns a pointer to the magic added.
5687 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5688 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5689 one instance of the same C<how>.
5691 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5692 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5693 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5694 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5696 (This is now used as a subroutine by C<sv_magic>.)
5701 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5702 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5706 PERL_ARGS_ASSERT_SV_MAGICEXT;
5708 SvUPGRADE(sv, SVt_PVMG);
5709 Newxz(mg, 1, MAGIC);
5710 mg->mg_moremagic = SvMAGIC(sv);
5711 SvMAGIC_set(sv, mg);
5713 /* Sometimes a magic contains a reference loop, where the sv and
5714 object refer to each other. To prevent a reference loop that
5715 would prevent such objects being freed, we look for such loops
5716 and if we find one we avoid incrementing the object refcount.
5718 Note we cannot do this to avoid self-tie loops as intervening RV must
5719 have its REFCNT incremented to keep it in existence.
5722 if (!obj || obj == sv ||
5723 how == PERL_MAGIC_arylen ||
5724 how == PERL_MAGIC_regdata ||
5725 how == PERL_MAGIC_regdatum ||
5726 how == PERL_MAGIC_symtab ||
5727 (SvTYPE(obj) == SVt_PVGV &&
5728 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5729 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5730 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5735 mg->mg_obj = SvREFCNT_inc_simple(obj);
5736 mg->mg_flags |= MGf_REFCOUNTED;
5739 /* Normal self-ties simply pass a null object, and instead of
5740 using mg_obj directly, use the SvTIED_obj macro to produce a
5741 new RV as needed. For glob "self-ties", we are tieing the PVIO
5742 with an RV obj pointing to the glob containing the PVIO. In
5743 this case, to avoid a reference loop, we need to weaken the
5747 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5748 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5754 mg->mg_len = namlen;
5757 mg->mg_ptr = savepvn(name, namlen);
5758 else if (namlen == HEf_SVKEY) {
5759 /* Yes, this is casting away const. This is only for the case of
5760 HEf_SVKEY. I think we need to document this aberation of the
5761 constness of the API, rather than making name non-const, as
5762 that change propagating outwards a long way. */
5763 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5765 mg->mg_ptr = (char *) name;
5767 mg->mg_virtual = (MGVTBL *) vtable;
5774 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5776 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5777 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5778 /* This sv is only a delegate. //g magic must be attached to
5783 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5784 &PL_vtbl_mglob, 0, 0);
5788 =for apidoc sv_magic
5790 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5791 necessary, then adds a new magic item of type C<how> to the head of the
5794 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5795 handling of the C<name> and C<namlen> arguments.
5797 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5798 to add more than one instance of the same C<how>.
5804 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5805 const char *const name, const I32 namlen)
5807 const MGVTBL *vtable;
5810 unsigned int vtable_index;
5812 PERL_ARGS_ASSERT_SV_MAGIC;
5814 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5815 || ((flags = PL_magic_data[how]),
5816 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5817 > magic_vtable_max))
5818 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5820 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5821 Useful for attaching extension internal data to perl vars.
5822 Note that multiple extensions may clash if magical scalars
5823 etc holding private data from one are passed to another. */
5825 vtable = (vtable_index == magic_vtable_max)
5826 ? NULL : PL_magic_vtables + vtable_index;
5828 if (SvREADONLY(sv)) {
5830 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5833 Perl_croak_no_modify();
5836 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5837 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5838 /* sv_magic() refuses to add a magic of the same 'how' as an
5841 if (how == PERL_MAGIC_taint)
5847 /* Force pos to be stored as characters, not bytes. */
5848 if (SvMAGICAL(sv) && DO_UTF8(sv)
5849 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5851 && mg->mg_flags & MGf_BYTES) {
5852 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5854 mg->mg_flags &= ~MGf_BYTES;
5857 /* Rest of work is done else where */
5858 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5861 case PERL_MAGIC_taint:
5864 case PERL_MAGIC_ext:
5865 case PERL_MAGIC_dbfile:
5872 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5879 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5881 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5882 for (mg = *mgp; mg; mg = *mgp) {
5883 const MGVTBL* const virt = mg->mg_virtual;
5884 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5885 *mgp = mg->mg_moremagic;
5886 if (virt && virt->svt_free)
5887 virt->svt_free(aTHX_ sv, mg);
5888 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5890 Safefree(mg->mg_ptr);
5891 else if (mg->mg_len == HEf_SVKEY)
5892 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5893 else if (mg->mg_type == PERL_MAGIC_utf8)
5894 Safefree(mg->mg_ptr);
5896 if (mg->mg_flags & MGf_REFCOUNTED)
5897 SvREFCNT_dec(mg->mg_obj);
5901 mgp = &mg->mg_moremagic;
5904 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5905 mg_magical(sv); /* else fix the flags now */
5914 =for apidoc sv_unmagic
5916 Removes all magic of type C<type> from an SV.
5922 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5924 PERL_ARGS_ASSERT_SV_UNMAGIC;
5925 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5929 =for apidoc sv_unmagicext
5931 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5937 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5939 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5940 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5944 =for apidoc sv_rvweaken
5946 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5947 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5948 push a back-reference to this RV onto the array of backreferences
5949 associated with that magic. If the RV is magical, set magic will be
5950 called after the RV is cleared. Silently ignores C<undef> and warns
5951 on already-weak references.
5957 Perl_sv_rvweaken(pTHX_ SV *const sv)
5961 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5963 if (!SvOK(sv)) /* let undefs pass */
5966 Perl_croak(aTHX_ "Can't weaken a nonreference");
5967 else if (SvWEAKREF(sv)) {
5968 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5971 else if (SvREADONLY(sv)) croak_no_modify();
5973 Perl_sv_add_backref(aTHX_ tsv, sv);
5975 SvREFCNT_dec_NN(tsv);
5980 =for apidoc sv_rvunweaken
5982 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5983 the backreference to this RV from the array of backreferences
5984 associated with the target SV, increment the refcount of the target.
5985 Silently ignores C<undef> and warns on non-weak references.
5991 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5995 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5997 if (!SvOK(sv)) /* let undefs pass */
6000 Perl_croak(aTHX_ "Can't unweaken a nonreference");
6001 else if (!SvWEAKREF(sv)) {
6002 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
6005 else if (SvREADONLY(sv)) croak_no_modify();
6010 SvREFCNT_inc_NN(tsv);
6011 Perl_sv_del_backref(aTHX_ tsv, sv);
6016 =for apidoc sv_get_backrefs
6018 If C<sv> is the target of a weak reference then it returns the back
6019 references structure associated with the sv; otherwise return C<NULL>.
6021 When returning a non-null result the type of the return is relevant. If it
6022 is an AV then the elements of the AV are the weak reference RVs which
6023 point at this item. If it is any other type then the item itself is the
6026 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6027 C<Perl_sv_kill_backrefs()>
6033 Perl_sv_get_backrefs(SV *const sv)
6037 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6039 /* find slot to store array or singleton backref */
6041 if (SvTYPE(sv) == SVt_PVHV) {
6043 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6044 backrefs = (SV *)iter->xhv_backreferences;
6046 } else if (SvMAGICAL(sv)) {
6047 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6049 backrefs = mg->mg_obj;
6054 /* Give tsv backref magic if it hasn't already got it, then push a
6055 * back-reference to sv onto the array associated with the backref magic.
6057 * As an optimisation, if there's only one backref and it's not an AV,
6058 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6059 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6063 /* A discussion about the backreferences array and its refcount:
6065 * The AV holding the backreferences is pointed to either as the mg_obj of
6066 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6067 * xhv_backreferences field. The array is created with a refcount
6068 * of 2. This means that if during global destruction the array gets
6069 * picked on before its parent to have its refcount decremented by the
6070 * random zapper, it won't actually be freed, meaning it's still there for
6071 * when its parent gets freed.
6073 * When the parent SV is freed, the extra ref is killed by
6074 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6075 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6077 * When a single backref SV is stored directly, it is not reference
6082 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6088 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6090 /* find slot to store array or singleton backref */
6092 if (SvTYPE(tsv) == SVt_PVHV) {
6093 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6096 mg = mg_find(tsv, PERL_MAGIC_backref);
6098 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6099 svp = &(mg->mg_obj);
6102 /* create or retrieve the array */
6104 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6105 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6109 mg->mg_flags |= MGf_REFCOUNTED;
6112 SvREFCNT_inc_simple_void_NN(av);
6113 /* av now has a refcnt of 2; see discussion above */
6114 av_extend(av, *svp ? 2 : 1);
6116 /* move single existing backref to the array */
6117 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6122 av = MUTABLE_AV(*svp);
6124 /* optimisation: store single backref directly in HvAUX or mg_obj */
6128 assert(SvTYPE(av) == SVt_PVAV);
6129 if (AvFILLp(av) >= AvMAX(av)) {
6130 av_extend(av, AvFILLp(av)+1);
6133 /* push new backref */
6134 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6137 /* delete a back-reference to ourselves from the backref magic associated
6138 * with the SV we point to.
6142 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6146 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6148 if (SvTYPE(tsv) == SVt_PVHV) {
6150 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6152 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6153 /* It's possible for the the last (strong) reference to tsv to have
6154 become freed *before* the last thing holding a weak reference.
6155 If both survive longer than the backreferences array, then when
6156 the referent's reference count drops to 0 and it is freed, it's
6157 not able to chase the backreferences, so they aren't NULLed.
6159 For example, a CV holds a weak reference to its stash. If both the
6160 CV and the stash survive longer than the backreferences array,
6161 and the CV gets picked for the SvBREAK() treatment first,
6162 *and* it turns out that the stash is only being kept alive because
6163 of an our variable in the pad of the CV, then midway during CV
6164 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6165 It ends up pointing to the freed HV. Hence it's chased in here, and
6166 if this block wasn't here, it would hit the !svp panic just below.
6168 I don't believe that "better" destruction ordering is going to help
6169 here - during global destruction there's always going to be the
6170 chance that something goes out of order. We've tried to make it
6171 foolproof before, and it only resulted in evolutionary pressure on
6172 fools. Which made us look foolish for our hubris. :-(
6178 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6179 svp = mg ? &(mg->mg_obj) : NULL;
6183 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6185 /* It's possible that sv is being freed recursively part way through the
6186 freeing of tsv. If this happens, the backreferences array of tsv has
6187 already been freed, and so svp will be NULL. If this is the case,
6188 we should not panic. Instead, nothing needs doing, so return. */
6189 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6191 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6192 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6195 if (SvTYPE(*svp) == SVt_PVAV) {
6199 AV * const av = (AV*)*svp;
6201 assert(!SvIS_FREED(av));
6205 /* for an SV with N weak references to it, if all those
6206 * weak refs are deleted, then sv_del_backref will be called
6207 * N times and O(N^2) compares will be done within the backref
6208 * array. To ameliorate this potential slowness, we:
6209 * 1) make sure this code is as tight as possible;
6210 * 2) when looking for SV, look for it at both the head and tail of the
6211 * array first before searching the rest, since some create/destroy
6212 * patterns will cause the backrefs to be freed in order.
6219 SV **p = &svp[fill];
6220 SV *const topsv = *p;
6227 /* We weren't the last entry.
6228 An unordered list has this property that you
6229 can take the last element off the end to fill
6230 the hole, and it's still an unordered list :-)
6236 break; /* should only be one */
6243 AvFILLp(av) = fill-1;
6245 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6246 /* freed AV; skip */
6249 /* optimisation: only a single backref, stored directly */
6251 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6252 (void*)*svp, (void*)sv);
6259 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6265 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6270 /* after multiple passes through Perl_sv_clean_all() for a thingy
6271 * that has badly leaked, the backref array may have gotten freed,
6272 * since we only protect it against 1 round of cleanup */
6273 if (SvIS_FREED(av)) {
6274 if (PL_in_clean_all) /* All is fair */
6277 "panic: magic_killbackrefs (freed backref AV/SV)");
6281 is_array = (SvTYPE(av) == SVt_PVAV);
6283 assert(!SvIS_FREED(av));
6286 last = svp + AvFILLp(av);
6289 /* optimisation: only a single backref, stored directly */
6295 while (svp <= last) {
6297 SV *const referrer = *svp;
6298 if (SvWEAKREF(referrer)) {
6299 /* XXX Should we check that it hasn't changed? */
6300 assert(SvROK(referrer));
6301 SvRV_set(referrer, 0);
6303 SvWEAKREF_off(referrer);
6304 SvSETMAGIC(referrer);
6305 } else if (SvTYPE(referrer) == SVt_PVGV ||
6306 SvTYPE(referrer) == SVt_PVLV) {
6307 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6308 /* You lookin' at me? */
6309 assert(GvSTASH(referrer));
6310 assert(GvSTASH(referrer) == (const HV *)sv);
6311 GvSTASH(referrer) = 0;
6312 } else if (SvTYPE(referrer) == SVt_PVCV ||
6313 SvTYPE(referrer) == SVt_PVFM) {
6314 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6315 /* You lookin' at me? */
6316 assert(CvSTASH(referrer));
6317 assert(CvSTASH(referrer) == (const HV *)sv);
6318 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6321 assert(SvTYPE(sv) == SVt_PVGV);
6322 /* You lookin' at me? */
6323 assert(CvGV(referrer));
6324 assert(CvGV(referrer) == (const GV *)sv);
6325 anonymise_cv_maybe(MUTABLE_GV(sv),
6326 MUTABLE_CV(referrer));
6331 "panic: magic_killbackrefs (flags=%" UVxf ")",
6332 (UV)SvFLAGS(referrer));
6343 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6349 =for apidoc sv_insert
6351 Inserts and/or replaces a string at the specified offset/length within the SV.
6352 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6353 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6354 C<offset>. Handles get magic.
6356 =for apidoc sv_insert_flags
6358 Same as C<sv_insert>, but the extra C<flags> are passed to the
6359 C<SvPV_force_flags> that applies to C<bigstr>.
6365 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6371 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6374 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6376 SvPV_force_flags(bigstr, curlen, flags);
6377 (void)SvPOK_only_UTF8(bigstr);
6379 if (little >= SvPVX(bigstr) &&
6380 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6381 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6382 or little...little+littlelen might overlap offset...offset+len we make a copy
6384 little = savepvn(little, littlelen);
6388 if (offset + len > curlen) {
6389 SvGROW(bigstr, offset+len+1);
6390 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6391 SvCUR_set(bigstr, offset+len);
6395 i = littlelen - len;
6396 if (i > 0) { /* string might grow */
6397 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6398 mid = big + offset + len;
6399 midend = bigend = big + SvCUR(bigstr);
6402 while (midend > mid) /* shove everything down */
6403 *--bigend = *--midend;
6404 Move(little,big+offset,littlelen,char);
6405 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6410 Move(little,SvPVX(bigstr)+offset,len,char);
6415 big = SvPVX(bigstr);
6418 bigend = big + SvCUR(bigstr);
6420 if (midend > bigend)
6421 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6424 if (mid - big > bigend - midend) { /* faster to shorten from end */
6426 Move(little, mid, littlelen,char);
6429 i = bigend - midend;
6431 Move(midend, mid, i,char);
6435 SvCUR_set(bigstr, mid - big);
6437 else if ((i = mid - big)) { /* faster from front */
6438 midend -= littlelen;
6440 Move(big, midend - i, i, char);
6441 sv_chop(bigstr,midend-i);
6443 Move(little, mid, littlelen,char);
6445 else if (littlelen) {
6446 midend -= littlelen;
6447 sv_chop(bigstr,midend);
6448 Move(little,midend,littlelen,char);
6451 sv_chop(bigstr,midend);
6457 =for apidoc sv_replace
6459 Make the first argument a copy of the second, then delete the original.
6460 The target SV physically takes over ownership of the body of the source SV
6461 and inherits its flags; however, the target keeps any magic it owns,
6462 and any magic in the source is discarded.
6463 Note that this is a rather specialist SV copying operation; most of the
6464 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6470 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6472 const U32 refcnt = SvREFCNT(sv);
6474 PERL_ARGS_ASSERT_SV_REPLACE;
6476 SV_CHECK_THINKFIRST_COW_DROP(sv);
6477 if (SvREFCNT(nsv) != 1) {
6478 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6479 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6481 if (SvMAGICAL(sv)) {
6485 sv_upgrade(nsv, SVt_PVMG);
6486 SvMAGIC_set(nsv, SvMAGIC(sv));
6487 SvFLAGS(nsv) |= SvMAGICAL(sv);
6489 SvMAGIC_set(sv, NULL);
6493 assert(!SvREFCNT(sv));
6494 #ifdef DEBUG_LEAKING_SCALARS
6495 sv->sv_flags = nsv->sv_flags;
6496 sv->sv_any = nsv->sv_any;
6497 sv->sv_refcnt = nsv->sv_refcnt;
6498 sv->sv_u = nsv->sv_u;
6500 StructCopy(nsv,sv,SV);
6502 if(SvTYPE(sv) == SVt_IV) {
6503 SET_SVANY_FOR_BODYLESS_IV(sv);
6507 SvREFCNT(sv) = refcnt;
6508 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6513 /* We're about to free a GV which has a CV that refers back to us.
6514 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6518 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6523 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6526 assert(SvREFCNT(gv) == 0);
6527 assert(isGV(gv) && isGV_with_GP(gv));
6529 assert(!CvANON(cv));
6530 assert(CvGV(cv) == gv);
6531 assert(!CvNAMED(cv));
6533 /* will the CV shortly be freed by gp_free() ? */
6534 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6535 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6539 /* if not, anonymise: */
6540 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6541 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6542 : newSVpvn_flags( "__ANON__", 8, 0 );
6543 sv_catpvs(gvname, "::__ANON__");
6544 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6545 SvREFCNT_dec_NN(gvname);
6549 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6554 =for apidoc sv_clear
6556 Clear an SV: call any destructors, free up any memory used by the body,
6557 and free the body itself. The SV's head is I<not> freed, although
6558 its type is set to all 1's so that it won't inadvertently be assumed
6559 to be live during global destruction etc.
6560 This function should only be called when C<REFCNT> is zero. Most of the time
6561 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6568 Perl_sv_clear(pTHX_ SV *const orig_sv)
6573 const struct body_details *sv_type_details;
6577 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6578 Not strictly necessary */
6580 PERL_ARGS_ASSERT_SV_CLEAR;
6582 /* within this loop, sv is the SV currently being freed, and
6583 * iter_sv is the most recent AV or whatever that's being iterated
6584 * over to provide more SVs */
6590 assert(SvREFCNT(sv) == 0);
6591 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6593 if (type <= SVt_IV) {
6594 /* See the comment in sv.h about the collusion between this
6595 * early return and the overloading of the NULL slots in the
6599 SvFLAGS(sv) &= SVf_BREAK;
6600 SvFLAGS(sv) |= SVTYPEMASK;
6604 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6605 for another purpose */
6606 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6608 if (type >= SVt_PVMG) {
6610 if (!curse(sv, 1)) goto get_next_sv;
6611 type = SvTYPE(sv); /* destructor may have changed it */
6613 /* Free back-references before magic, in case the magic calls
6614 * Perl code that has weak references to sv. */
6615 if (type == SVt_PVHV) {
6616 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6620 else if (SvMAGIC(sv)) {
6621 /* Free back-references before other types of magic. */
6622 sv_unmagic(sv, PERL_MAGIC_backref);
6628 /* case SVt_INVLIST: */
6631 IoIFP(sv) != PerlIO_stdin() &&
6632 IoIFP(sv) != PerlIO_stdout() &&
6633 IoIFP(sv) != PerlIO_stderr() &&
6634 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6636 io_close(MUTABLE_IO(sv), NULL, FALSE,
6637 (IoTYPE(sv) == IoTYPE_WRONLY ||
6638 IoTYPE(sv) == IoTYPE_RDWR ||
6639 IoTYPE(sv) == IoTYPE_APPEND));
6641 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6642 PerlDir_close(IoDIRP(sv));
6643 IoDIRP(sv) = (DIR*)NULL;
6644 Safefree(IoTOP_NAME(sv));
6645 Safefree(IoFMT_NAME(sv));
6646 Safefree(IoBOTTOM_NAME(sv));
6647 if ((const GV *)sv == PL_statgv)
6651 /* FIXME for plugins */
6652 pregfree2((REGEXP*) sv);
6656 cv_undef(MUTABLE_CV(sv));
6657 /* If we're in a stash, we don't own a reference to it.
6658 * However it does have a back reference to us, which needs to
6660 if ((stash = CvSTASH(sv)))
6661 sv_del_backref(MUTABLE_SV(stash), sv);
6664 if (PL_last_swash_hv == (const HV *)sv) {
6665 PL_last_swash_hv = NULL;
6667 if (HvTOTALKEYS((HV*)sv) > 0) {
6669 /* this statement should match the one at the beginning of
6670 * hv_undef_flags() */
6671 if ( PL_phase != PERL_PHASE_DESTRUCT
6672 && (hek = HvNAME_HEK((HV*)sv)))
6674 if (PL_stashcache) {
6675 DEBUG_o(Perl_deb(aTHX_
6676 "sv_clear clearing PL_stashcache for '%" HEKf
6679 (void)hv_deletehek(PL_stashcache,
6682 hv_name_set((HV*)sv, NULL, 0, 0);
6685 /* save old iter_sv in unused SvSTASH field */
6686 assert(!SvOBJECT(sv));
6687 SvSTASH(sv) = (HV*)iter_sv;
6690 /* save old hash_index in unused SvMAGIC field */
6691 assert(!SvMAGICAL(sv));
6692 assert(!SvMAGIC(sv));
6693 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6696 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6697 goto get_next_sv; /* process this new sv */
6699 /* free empty hash */
6700 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6701 assert(!HvARRAY((HV*)sv));
6705 AV* av = MUTABLE_AV(sv);
6706 if (PL_comppad == av) {
6710 if (AvREAL(av) && AvFILLp(av) > -1) {
6711 next_sv = AvARRAY(av)[AvFILLp(av)--];
6712 /* save old iter_sv in top-most slot of AV,
6713 * and pray that it doesn't get wiped in the meantime */
6714 AvARRAY(av)[AvMAX(av)] = iter_sv;
6716 goto get_next_sv; /* process this new sv */
6718 Safefree(AvALLOC(av));
6723 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6724 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6725 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6726 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6728 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6729 SvREFCNT_dec(LvTARG(sv));
6731 /* SvLEN points to a regex body. Free the body, then
6732 * set SvLEN to whatever value was in the now-freed
6733 * regex body. The PVX buffer is shared by multiple re's
6734 * and only freed once, by the re whose len in non-null */
6735 STRLEN len = ReANY(sv)->xpv_len;
6736 pregfree2((REGEXP*) sv);
6737 SvLEN_set((sv), len);
6742 if (isGV_with_GP(sv)) {
6743 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6744 && HvENAME_get(stash))
6745 mro_method_changed_in(stash);
6746 gp_free(MUTABLE_GV(sv));
6748 unshare_hek(GvNAME_HEK(sv));
6749 /* If we're in a stash, we don't own a reference to it.
6750 * However it does have a back reference to us, which
6751 * needs to be cleared. */
6752 if ((stash = GvSTASH(sv)))
6753 sv_del_backref(MUTABLE_SV(stash), sv);
6755 /* FIXME. There are probably more unreferenced pointers to SVs
6756 * in the interpreter struct that we should check and tidy in
6757 * a similar fashion to this: */
6758 /* See also S_sv_unglob, which does the same thing. */
6759 if ((const GV *)sv == PL_last_in_gv)
6760 PL_last_in_gv = NULL;
6761 else if ((const GV *)sv == PL_statgv)
6763 else if ((const GV *)sv == PL_stderrgv)
6772 /* Don't bother with SvOOK_off(sv); as we're only going to
6776 SvOOK_offset(sv, offset);
6777 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6778 /* Don't even bother with turning off the OOK flag. */
6783 SV * const target = SvRV(sv);
6785 sv_del_backref(target, sv);
6791 else if (SvPVX_const(sv)
6792 && !(SvTYPE(sv) == SVt_PVIO
6793 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6798 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6803 if (CowREFCNT(sv)) {
6810 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6815 Safefree(SvPVX_mutable(sv));
6819 else if (SvPVX_const(sv) && SvLEN(sv)
6820 && !(SvTYPE(sv) == SVt_PVIO
6821 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6822 Safefree(SvPVX_mutable(sv));
6823 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6824 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6834 SvFLAGS(sv) &= SVf_BREAK;
6835 SvFLAGS(sv) |= SVTYPEMASK;
6837 sv_type_details = bodies_by_type + type;
6838 if (sv_type_details->arena) {
6839 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6840 &PL_body_roots[type]);
6842 else if (sv_type_details->body_size) {
6843 safefree(SvANY(sv));
6847 /* caller is responsible for freeing the head of the original sv */
6848 if (sv != orig_sv && !SvREFCNT(sv))
6851 /* grab and free next sv, if any */
6859 else if (!iter_sv) {
6861 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6862 AV *const av = (AV*)iter_sv;
6863 if (AvFILLp(av) > -1) {
6864 sv = AvARRAY(av)[AvFILLp(av)--];
6866 else { /* no more elements of current AV to free */
6869 /* restore previous value, squirrelled away */
6870 iter_sv = AvARRAY(av)[AvMAX(av)];
6871 Safefree(AvALLOC(av));
6874 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6875 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6876 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6877 /* no more elements of current HV to free */
6880 /* Restore previous values of iter_sv and hash_index,
6881 * squirrelled away */
6882 assert(!SvOBJECT(sv));
6883 iter_sv = (SV*)SvSTASH(sv);
6884 assert(!SvMAGICAL(sv));
6885 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6887 /* perl -DA does not like rubbish in SvMAGIC. */
6891 /* free any remaining detritus from the hash struct */
6892 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6893 assert(!HvARRAY((HV*)sv));
6898 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6902 if (!SvREFCNT(sv)) {
6906 if (--(SvREFCNT(sv)))
6910 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6911 "Attempt to free temp prematurely: SV 0x%" UVxf
6912 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6916 if (SvIMMORTAL(sv)) {
6917 /* make sure SvREFCNT(sv)==0 happens very seldom */
6918 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6927 /* This routine curses the sv itself, not the object referenced by sv. So
6928 sv does not have to be ROK. */
6931 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6932 PERL_ARGS_ASSERT_CURSE;
6933 assert(SvOBJECT(sv));
6935 if (PL_defstash && /* Still have a symbol table? */
6941 stash = SvSTASH(sv);
6942 assert(SvTYPE(stash) == SVt_PVHV);
6943 if (HvNAME(stash)) {
6944 CV* destructor = NULL;
6945 struct mro_meta *meta;
6947 assert (SvOOK(stash));
6949 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6952 /* don't make this an initialization above the assert, since it needs
6954 meta = HvMROMETA(stash);
6955 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6956 destructor = meta->destroy;
6957 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6958 (void *)destructor, HvNAME(stash)) );
6961 bool autoload = FALSE;
6963 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6965 destructor = GvCV(gv);
6967 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6968 GV_AUTOLOAD_ISMETHOD);
6970 destructor = GvCV(gv);
6974 /* we don't cache AUTOLOAD for DESTROY, since this code
6975 would then need to set $__PACKAGE__::AUTOLOAD, or the
6976 equivalent for XS AUTOLOADs */
6978 meta->destroy_gen = PL_sub_generation;
6979 meta->destroy = destructor;
6981 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6982 (void *)destructor, HvNAME(stash)) );
6985 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6989 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6991 /* A constant subroutine can have no side effects, so
6992 don't bother calling it. */
6993 && !CvCONST(destructor)
6994 /* Don't bother calling an empty destructor or one that
6995 returns immediately. */
6996 && (CvISXSUB(destructor)
6997 || (CvSTART(destructor)
6998 && (CvSTART(destructor)->op_next->op_type
7000 && (CvSTART(destructor)->op_next->op_type
7002 || CvSTART(destructor)->op_next->op_next->op_type
7008 SV* const tmpref = newRV(sv);
7009 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
7011 PUSHSTACKi(PERLSI_DESTROY);
7016 call_sv(MUTABLE_SV(destructor),
7017 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
7021 if(SvREFCNT(tmpref) < 2) {
7022 /* tmpref is not kept alive! */
7024 SvRV_set(tmpref, NULL);
7027 SvREFCNT_dec_NN(tmpref);
7030 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7033 if (check_refcnt && SvREFCNT(sv)) {
7034 if (PL_in_clean_objs)
7036 "DESTROY created new reference to dead object '%" HEKf "'",
7037 HEKfARG(HvNAME_HEK(stash)));
7038 /* DESTROY gave object new lease on life */
7044 HV * const stash = SvSTASH(sv);
7045 /* Curse before freeing the stash, as freeing the stash could cause
7046 a recursive call into S_curse. */
7047 SvOBJECT_off(sv); /* Curse the object. */
7048 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7049 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7055 =for apidoc sv_newref
7057 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7064 Perl_sv_newref(pTHX_ SV *const sv)
7066 PERL_UNUSED_CONTEXT;
7075 Decrement an SV's reference count, and if it drops to zero, call
7076 C<sv_clear> to invoke destructors and free up any memory used by
7077 the body; finally, deallocating the SV's head itself.
7078 Normally called via a wrapper macro C<SvREFCNT_dec>.
7084 Perl_sv_free(pTHX_ SV *const sv)
7090 /* Private helper function for SvREFCNT_dec().
7091 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7094 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7098 PERL_ARGS_ASSERT_SV_FREE2;
7100 if (LIKELY( rc == 1 )) {
7106 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7107 "Attempt to free temp prematurely: SV 0x%" UVxf
7108 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7112 if (SvIMMORTAL(sv)) {
7113 /* make sure SvREFCNT(sv)==0 happens very seldom */
7114 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7118 if (! SvREFCNT(sv)) /* may have have been resurrected */
7123 /* handle exceptional cases */
7127 if (SvFLAGS(sv) & SVf_BREAK)
7128 /* this SV's refcnt has been artificially decremented to
7129 * trigger cleanup */
7131 if (PL_in_clean_all) /* All is fair */
7133 if (SvIMMORTAL(sv)) {
7134 /* make sure SvREFCNT(sv)==0 happens very seldom */
7135 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7138 if (ckWARN_d(WARN_INTERNAL)) {
7139 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7140 Perl_dump_sv_child(aTHX_ sv);
7142 #ifdef DEBUG_LEAKING_SCALARS
7145 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7146 if (PL_warnhook == PERL_WARNHOOK_FATAL
7147 || ckDEAD(packWARN(WARN_INTERNAL))) {
7148 /* Don't let Perl_warner cause us to escape our fate: */
7152 /* This may not return: */
7153 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7154 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7155 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7158 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7168 Returns the length of the string in the SV. Handles magic and type
7169 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7170 gives raw access to the C<xpv_cur> slot.
7176 Perl_sv_len(pTHX_ SV *const sv)
7183 (void)SvPV_const(sv, len);
7188 =for apidoc sv_len_utf8
7190 Returns the number of characters in the string in an SV, counting wide
7191 UTF-8 bytes as a single character. Handles magic and type coercion.
7197 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7198 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7199 * (Note that the mg_len is not the length of the mg_ptr field.
7200 * This allows the cache to store the character length of the string without
7201 * needing to malloc() extra storage to attach to the mg_ptr.)
7206 Perl_sv_len_utf8(pTHX_ SV *const sv)
7212 return sv_len_utf8_nomg(sv);
7216 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7219 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7221 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7223 if (PL_utf8cache && SvUTF8(sv)) {
7225 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7227 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7228 if (mg->mg_len != -1)
7231 /* We can use the offset cache for a headstart.
7232 The longer value is stored in the first pair. */
7233 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7235 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7239 if (PL_utf8cache < 0) {
7240 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7241 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7245 ulen = Perl_utf8_length(aTHX_ s, s + len);
7246 utf8_mg_len_cache_update(sv, &mg, ulen);
7250 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7253 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7256 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7257 STRLEN *const uoffset_p, bool *const at_end)
7259 const U8 *s = start;
7260 STRLEN uoffset = *uoffset_p;
7262 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7264 while (s < send && uoffset) {
7271 else if (s > send) {
7273 /* This is the existing behaviour. Possibly it should be a croak, as
7274 it's actually a bounds error */
7277 *uoffset_p -= uoffset;
7281 /* Given the length of the string in both bytes and UTF-8 characters, decide
7282 whether to walk forwards or backwards to find the byte corresponding to
7283 the passed in UTF-8 offset. */
7285 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7286 STRLEN uoffset, const STRLEN uend)
7288 STRLEN backw = uend - uoffset;
7290 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7292 if (uoffset < 2 * backw) {
7293 /* The assumption is that going forwards is twice the speed of going
7294 forward (that's where the 2 * backw comes from).
7295 (The real figure of course depends on the UTF-8 data.) */
7296 const U8 *s = start;
7298 while (s < send && uoffset--)
7308 while (UTF8_IS_CONTINUATION(*send))
7311 return send - start;
7314 /* For the string representation of the given scalar, find the byte
7315 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7316 give another position in the string, *before* the sought offset, which
7317 (which is always true, as 0, 0 is a valid pair of positions), which should
7318 help reduce the amount of linear searching.
7319 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7320 will be used to reduce the amount of linear searching. The cache will be
7321 created if necessary, and the found value offered to it for update. */
7323 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7324 const U8 *const send, STRLEN uoffset,
7325 STRLEN uoffset0, STRLEN boffset0)
7327 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7329 bool at_end = FALSE;
7331 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7333 assert (uoffset >= uoffset0);
7338 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7340 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7341 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7342 if ((*mgp)->mg_ptr) {
7343 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7344 if (cache[0] == uoffset) {
7345 /* An exact match. */
7348 if (cache[2] == uoffset) {
7349 /* An exact match. */
7353 if (cache[0] < uoffset) {
7354 /* The cache already knows part of the way. */
7355 if (cache[0] > uoffset0) {
7356 /* The cache knows more than the passed in pair */
7357 uoffset0 = cache[0];
7358 boffset0 = cache[1];
7360 if ((*mgp)->mg_len != -1) {
7361 /* And we know the end too. */
7363 + sv_pos_u2b_midway(start + boffset0, send,
7365 (*mgp)->mg_len - uoffset0);
7367 uoffset -= uoffset0;
7369 + sv_pos_u2b_forwards(start + boffset0,
7370 send, &uoffset, &at_end);
7371 uoffset += uoffset0;
7374 else if (cache[2] < uoffset) {
7375 /* We're between the two cache entries. */
7376 if (cache[2] > uoffset0) {
7377 /* and the cache knows more than the passed in pair */
7378 uoffset0 = cache[2];
7379 boffset0 = cache[3];
7383 + sv_pos_u2b_midway(start + boffset0,
7386 cache[0] - uoffset0);
7389 + sv_pos_u2b_midway(start + boffset0,
7392 cache[2] - uoffset0);
7396 else if ((*mgp)->mg_len != -1) {
7397 /* If we can take advantage of a passed in offset, do so. */
7398 /* In fact, offset0 is either 0, or less than offset, so don't
7399 need to worry about the other possibility. */
7401 + sv_pos_u2b_midway(start + boffset0, send,
7403 (*mgp)->mg_len - uoffset0);
7408 if (!found || PL_utf8cache < 0) {
7409 STRLEN real_boffset;
7410 uoffset -= uoffset0;
7411 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7412 send, &uoffset, &at_end);
7413 uoffset += uoffset0;
7415 if (found && PL_utf8cache < 0)
7416 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7418 boffset = real_boffset;
7421 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7423 utf8_mg_len_cache_update(sv, mgp, uoffset);
7425 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7432 =for apidoc sv_pos_u2b_flags
7434 Converts the offset from a count of UTF-8 chars from
7435 the start of the string, to a count of the equivalent number of bytes; if
7436 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7437 C<offset>, rather than from the start
7438 of the string. Handles type coercion.
7439 C<flags> is passed to C<SvPV_flags>, and usually should be
7440 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7446 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7447 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7448 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7453 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7460 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7462 start = (U8*)SvPV_flags(sv, len, flags);
7464 const U8 * const send = start + len;
7466 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7469 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7470 is 0, and *lenp is already set to that. */) {
7471 /* Convert the relative offset to absolute. */
7472 const STRLEN uoffset2 = uoffset + *lenp;
7473 const STRLEN boffset2
7474 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7475 uoffset, boffset) - boffset;
7489 =for apidoc sv_pos_u2b
7491 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7492 the start of the string, to a count of the equivalent number of bytes; if
7493 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7494 the offset, rather than from the start of the string. Handles magic and
7497 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7504 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7505 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7506 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7510 /* This function is subject to size and sign problems */
7513 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7515 PERL_ARGS_ASSERT_SV_POS_U2B;
7518 STRLEN ulen = (STRLEN)*lenp;
7519 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7520 SV_GMAGIC|SV_CONST_RETURN);
7523 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7524 SV_GMAGIC|SV_CONST_RETURN);
7529 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7532 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7533 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7536 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7537 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7538 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7542 (*mgp)->mg_len = ulen;
7545 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7546 byte length pairing. The (byte) length of the total SV is passed in too,
7547 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7548 may not have updated SvCUR, so we can't rely on reading it directly.
7550 The proffered utf8/byte length pairing isn't used if the cache already has
7551 two pairs, and swapping either for the proffered pair would increase the
7552 RMS of the intervals between known byte offsets.
7554 The cache itself consists of 4 STRLEN values
7555 0: larger UTF-8 offset
7556 1: corresponding byte offset
7557 2: smaller UTF-8 offset
7558 3: corresponding byte offset
7560 Unused cache pairs have the value 0, 0.
7561 Keeping the cache "backwards" means that the invariant of
7562 cache[0] >= cache[2] is maintained even with empty slots, which means that
7563 the code that uses it doesn't need to worry if only 1 entry has actually
7564 been set to non-zero. It also makes the "position beyond the end of the
7565 cache" logic much simpler, as the first slot is always the one to start
7569 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7570 const STRLEN utf8, const STRLEN blen)
7574 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7579 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7580 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7581 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7583 (*mgp)->mg_len = -1;
7587 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7588 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7589 (*mgp)->mg_ptr = (char *) cache;
7593 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7594 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7595 a pointer. Note that we no longer cache utf8 offsets on refer-
7596 ences, but this check is still a good idea, for robustness. */
7597 const U8 *start = (const U8 *) SvPVX_const(sv);
7598 const STRLEN realutf8 = utf8_length(start, start + byte);
7600 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7604 /* Cache is held with the later position first, to simplify the code
7605 that deals with unbounded ends. */
7607 ASSERT_UTF8_CACHE(cache);
7608 if (cache[1] == 0) {
7609 /* Cache is totally empty */
7612 } else if (cache[3] == 0) {
7613 if (byte > cache[1]) {
7614 /* New one is larger, so goes first. */
7615 cache[2] = cache[0];
7616 cache[3] = cache[1];
7624 /* float casts necessary? XXX */
7625 #define THREEWAY_SQUARE(a,b,c,d) \
7626 ((float)((d) - (c))) * ((float)((d) - (c))) \
7627 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7628 + ((float)((b) - (a))) * ((float)((b) - (a)))
7630 /* Cache has 2 slots in use, and we know three potential pairs.
7631 Keep the two that give the lowest RMS distance. Do the
7632 calculation in bytes simply because we always know the byte
7633 length. squareroot has the same ordering as the positive value,
7634 so don't bother with the actual square root. */
7635 if (byte > cache[1]) {
7636 /* New position is after the existing pair of pairs. */
7637 const float keep_earlier
7638 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7639 const float keep_later
7640 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7642 if (keep_later < keep_earlier) {
7643 cache[2] = cache[0];
7644 cache[3] = cache[1];
7650 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7651 float b, c, keep_earlier;
7652 if (byte > cache[3]) {
7653 /* New position is between the existing pair of pairs. */
7654 b = (float)cache[3];
7657 /* New position is before the existing pair of pairs. */
7659 c = (float)cache[3];
7661 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7662 if (byte > cache[3]) {
7663 if (keep_later < keep_earlier) {
7673 if (! (keep_later < keep_earlier)) {
7674 cache[0] = cache[2];
7675 cache[1] = cache[3];
7682 ASSERT_UTF8_CACHE(cache);
7685 /* We already know all of the way, now we may be able to walk back. The same
7686 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7687 backward is half the speed of walking forward. */
7689 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7690 const U8 *end, STRLEN endu)
7692 const STRLEN forw = target - s;
7693 STRLEN backw = end - target;
7695 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7697 if (forw < 2 * backw) {
7698 return utf8_length(s, target);
7701 while (end > target) {
7703 while (UTF8_IS_CONTINUATION(*end)) {
7712 =for apidoc sv_pos_b2u_flags
7714 Converts C<offset> from a count of bytes from the start of the string, to
7715 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7716 C<flags> is passed to C<SvPV_flags>, and usually should be
7717 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7723 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7724 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7729 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7732 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7738 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7740 s = (const U8*)SvPV_flags(sv, blen, flags);
7743 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7744 ", byte=%" UVuf, (UV)blen, (UV)offset);
7750 && SvTYPE(sv) >= SVt_PVMG
7751 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7754 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7755 if (cache[1] == offset) {
7756 /* An exact match. */
7759 if (cache[3] == offset) {
7760 /* An exact match. */
7764 if (cache[1] < offset) {
7765 /* We already know part of the way. */
7766 if (mg->mg_len != -1) {
7767 /* Actually, we know the end too. */
7769 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7770 s + blen, mg->mg_len - cache[0]);
7772 len = cache[0] + utf8_length(s + cache[1], send);
7775 else if (cache[3] < offset) {
7776 /* We're between the two cached pairs, so we do the calculation
7777 offset by the byte/utf-8 positions for the earlier pair,
7778 then add the utf-8 characters from the string start to
7780 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7781 s + cache[1], cache[0] - cache[2])
7785 else { /* cache[3] > offset */
7786 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7790 ASSERT_UTF8_CACHE(cache);
7792 } else if (mg->mg_len != -1) {
7793 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7797 if (!found || PL_utf8cache < 0) {
7798 const STRLEN real_len = utf8_length(s, send);
7800 if (found && PL_utf8cache < 0)
7801 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7807 utf8_mg_len_cache_update(sv, &mg, len);
7809 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7816 =for apidoc sv_pos_b2u
7818 Converts the value pointed to by C<offsetp> from a count of bytes from the
7819 start of the string, to a count of the equivalent number of UTF-8 chars.
7820 Handles magic and type coercion.
7822 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7829 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7830 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7835 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7837 PERL_ARGS_ASSERT_SV_POS_B2U;
7842 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7843 SV_GMAGIC|SV_CONST_RETURN);
7847 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7848 STRLEN real, SV *const sv)
7850 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7852 /* As this is debugging only code, save space by keeping this test here,
7853 rather than inlining it in all the callers. */
7854 if (from_cache == real)
7857 /* Need to turn the assertions off otherwise we may recurse infinitely
7858 while printing error messages. */
7859 SAVEI8(PL_utf8cache);
7861 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7862 func, (UV) from_cache, (UV) real, SVfARG(sv));
7868 Returns a boolean indicating whether the strings in the two SVs are
7869 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7870 coerce its args to strings if necessary.
7872 =for apidoc sv_eq_flags
7874 Returns a boolean indicating whether the strings in the two SVs are
7875 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7876 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7882 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7894 /* if pv1 and pv2 are the same, second SvPV_const call may
7895 * invalidate pv1 (if we are handling magic), so we may need to
7897 if (sv1 == sv2 && flags & SV_GMAGIC
7898 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7899 pv1 = SvPV_const(sv1, cur1);
7900 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7902 pv1 = SvPV_flags_const(sv1, cur1, flags);
7910 pv2 = SvPV_flags_const(sv2, cur2, flags);
7912 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7913 /* Differing utf8ness. */
7915 /* sv1 is the UTF-8 one */
7916 return bytes_cmp_utf8((const U8*)pv2, cur2,
7917 (const U8*)pv1, cur1) == 0;
7920 /* sv2 is the UTF-8 one */
7921 return bytes_cmp_utf8((const U8*)pv1, cur1,
7922 (const U8*)pv2, cur2) == 0;
7927 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7935 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7936 string in C<sv1> is less than, equal to, or greater than the string in
7937 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7938 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7940 =for apidoc sv_cmp_flags
7942 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7943 string in C<sv1> is less than, equal to, or greater than the string in
7944 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7945 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7946 also C<L</sv_cmp_locale_flags>>.
7952 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7954 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7958 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7962 const char *pv1, *pv2;
7964 SV *svrecode = NULL;
7971 pv1 = SvPV_flags_const(sv1, cur1, flags);
7978 pv2 = SvPV_flags_const(sv2, cur2, flags);
7980 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7981 /* Differing utf8ness. */
7983 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7984 (const U8*)pv1, cur1);
7985 return retval ? retval < 0 ? -1 : +1 : 0;
7988 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7989 (const U8*)pv2, cur2);
7990 return retval ? retval < 0 ? -1 : +1 : 0;
7994 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7997 cmp = cur2 ? -1 : 0;
8001 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
8004 if (! DO_UTF8(sv1)) {
8006 const I32 retval = memcmp((const void*)pv1,
8010 cmp = retval < 0 ? -1 : 1;
8011 } else if (cur1 == cur2) {
8014 cmp = cur1 < cur2 ? -1 : 1;
8018 else { /* Both are to be treated as UTF-EBCDIC */
8020 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
8021 * which remaps code points 0-255. We therefore generally have to
8022 * unmap back to the original values to get an accurate comparison.
8023 * But we don't have to do that for UTF-8 invariants, as by
8024 * definition, they aren't remapped, nor do we have to do it for
8025 * above-latin1 code points, as they also aren't remapped. (This
8026 * code also works on ASCII platforms, but the memcmp() above is
8029 const char *e = pv1 + shortest_len;
8031 /* Find the first bytes that differ between the two strings */
8032 while (pv1 < e && *pv1 == *pv2) {
8038 if (pv1 == e) { /* Are the same all the way to the end */
8042 cmp = cur1 < cur2 ? -1 : 1;
8045 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8046 * in the strings were. The current bytes may or may not be
8047 * at the beginning of a character. But neither or both are
8048 * (or else earlier bytes would have been different). And
8049 * if we are in the middle of a character, the two
8050 * characters are comprised of the same number of bytes
8051 * (because in this case the start bytes are the same, and
8052 * the start bytes encode the character's length). */
8053 if (UTF8_IS_INVARIANT(*pv1))
8055 /* If both are invariants; can just compare directly */
8056 if (UTF8_IS_INVARIANT(*pv2)) {
8057 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8059 else /* Since *pv1 is invariant, it is the whole character,
8060 which means it is at the beginning of a character.
8061 That means pv2 is also at the beginning of a
8062 character (see earlier comment). Since it isn't
8063 invariant, it must be a start byte. If it starts a
8064 character whose code point is above 255, that
8065 character is greater than any single-byte char, which
8067 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8072 /* Here, pv2 points to a character composed of 2 bytes
8073 * whose code point is < 256. Get its code point and
8074 * compare with *pv1 */
8075 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8080 else /* The code point starting at pv1 isn't a single byte */
8081 if (UTF8_IS_INVARIANT(*pv2))
8083 /* But here, the code point starting at *pv2 is a single byte,
8084 * and so *pv1 must begin a character, hence is a start byte.
8085 * If that character is above 255, it is larger than any
8086 * single-byte char, which *pv2 is */
8087 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8091 /* Here, pv1 points to a character composed of 2 bytes
8092 * whose code point is < 256. Get its code point and
8093 * compare with the single byte character *pv2 */
8094 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8099 else /* Here, we've ruled out either *pv1 and *pv2 being
8100 invariant. That means both are part of variants, but not
8101 necessarily at the start of a character */
8102 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8103 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8105 /* Here, at least one is the start of a character, which means
8106 * the other is also a start byte. And the code point of at
8107 * least one of the characters is above 255. It is a
8108 * characteristic of UTF-EBCDIC that all start bytes for
8109 * above-latin1 code points are well behaved as far as code
8110 * point comparisons go, and all are larger than all other
8111 * start bytes, so the comparison with those is also well
8113 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8116 /* Here both *pv1 and *pv2 are part of variant characters.
8117 * They could be both continuations, or both start characters.
8118 * (One or both could even be an illegal start character (for
8119 * an overlong) which for the purposes of sorting we treat as
8121 if (UTF8_IS_CONTINUATION(*pv1)) {
8123 /* If they are continuations for code points above 255,
8124 * then comparing the current byte is sufficient, as there
8125 * is no remapping of these and so the comparison is
8126 * well-behaved. We determine if they are such
8127 * continuations by looking at the preceding byte. It
8128 * could be a start byte, from which we can tell if it is
8129 * for an above 255 code point. Or it could be a
8130 * continuation, which means the character occupies at
8131 * least 3 bytes, so must be above 255. */
8132 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8133 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8135 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8139 /* Here, the continuations are for code points below 256;
8140 * back up one to get to the start byte */
8145 /* We need to get the actual native code point of each of these
8146 * variants in order to compare them */
8147 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8148 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8157 SvREFCNT_dec(svrecode);
8163 =for apidoc sv_cmp_locale
8165 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8166 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8167 if necessary. See also C<L</sv_cmp>>.
8169 =for apidoc sv_cmp_locale_flags
8171 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8172 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8173 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8174 C<L</sv_cmp_flags>>.
8180 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8182 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8186 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8189 #ifdef USE_LOCALE_COLLATE
8195 if (PL_collation_standard)
8200 /* Revert to using raw compare if both operands exist, but either one
8201 * doesn't transform properly for collation */
8203 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8207 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8213 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8214 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8217 if (!pv1 || !len1) {
8228 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8231 return retval < 0 ? -1 : 1;
8234 * When the result of collation is equality, that doesn't mean
8235 * that there are no differences -- some locales exclude some
8236 * characters from consideration. So to avoid false equalities,
8237 * we use the raw string as a tiebreaker.
8244 PERL_UNUSED_ARG(flags);
8245 #endif /* USE_LOCALE_COLLATE */
8247 return sv_cmp(sv1, sv2);
8251 #ifdef USE_LOCALE_COLLATE
8254 =for apidoc sv_collxfrm
8256 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8257 C<L</sv_collxfrm_flags>>.
8259 =for apidoc sv_collxfrm_flags
8261 Add Collate Transform magic to an SV if it doesn't already have it. If the
8262 flags contain C<SV_GMAGIC>, it handles get-magic.
8264 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8265 scalar data of the variable, but transformed to such a format that a normal
8266 memory comparison can be used to compare the data according to the locale
8273 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8277 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8279 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8281 /* If we don't have collation magic on 'sv', or the locale has changed
8282 * since the last time we calculated it, get it and save it now */
8283 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8288 /* Free the old space */
8290 Safefree(mg->mg_ptr);
8292 s = SvPV_flags_const(sv, len, flags);
8293 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8295 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8310 if (mg && mg->mg_ptr) {
8312 return mg->mg_ptr + sizeof(PL_collation_ix);
8320 #endif /* USE_LOCALE_COLLATE */
8323 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8325 SV * const tsv = newSV(0);
8328 sv_gets(tsv, fp, 0);
8329 sv_utf8_upgrade_nomg(tsv);
8330 SvCUR_set(sv,append);
8333 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8337 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8340 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8341 /* Grab the size of the record we're getting */
8342 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8349 /* With a true, record-oriented file on VMS, we need to use read directly
8350 * to ensure that we respect RMS record boundaries. The user is responsible
8351 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8352 * record size) field. N.B. This is likely to produce invalid results on
8353 * varying-width character data when a record ends mid-character.
8355 fd = PerlIO_fileno(fp);
8357 && PerlLIO_fstat(fd, &st) == 0
8358 && (st.st_fab_rfm == FAB$C_VAR
8359 || st.st_fab_rfm == FAB$C_VFC
8360 || st.st_fab_rfm == FAB$C_FIX)) {
8362 bytesread = PerlLIO_read(fd, buffer, recsize);
8364 else /* in-memory file from PerlIO::Scalar
8365 * or not a record-oriented file
8369 bytesread = PerlIO_read(fp, buffer, recsize);
8371 /* At this point, the logic in sv_get() means that sv will
8372 be treated as utf-8 if the handle is utf8.
8374 if (PerlIO_isutf8(fp) && bytesread > 0) {
8375 char *bend = buffer + bytesread;
8376 char *bufp = buffer;
8377 size_t charcount = 0;
8378 bool charstart = TRUE;
8381 while (charcount < recsize) {
8382 /* count accumulated characters */
8383 while (bufp < bend) {
8385 skip = UTF8SKIP(bufp);
8387 if (bufp + skip > bend) {
8388 /* partial at the end */
8399 if (charcount < recsize) {
8401 STRLEN bufp_offset = bufp - buffer;
8402 SSize_t morebytesread;
8404 /* originally I read enough to fill any incomplete
8405 character and the first byte of the next
8406 character if needed, but if there's many
8407 multi-byte encoded characters we're going to be
8408 making a read call for every character beyond
8409 the original read size.
8411 So instead, read the rest of the character if
8412 any, and enough bytes to match at least the
8413 start bytes for each character we're going to
8417 readsize = recsize - charcount;
8419 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8420 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8421 bend = buffer + bytesread;
8422 morebytesread = PerlIO_read(fp, bend, readsize);
8423 if (morebytesread <= 0) {
8424 /* we're done, if we still have incomplete
8425 characters the check code in sv_gets() will
8428 I'd originally considered doing
8429 PerlIO_ungetc() on all but the lead
8430 character of the incomplete character, but
8431 read() doesn't do that, so I don't.
8436 /* prepare to scan some more */
8437 bytesread += morebytesread;
8438 bend = buffer + bytesread;
8439 bufp = buffer + bufp_offset;
8447 SvCUR_set(sv, bytesread + append);
8448 buffer[bytesread] = '\0';
8449 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8455 Get a line from the filehandle and store it into the SV, optionally
8456 appending to the currently-stored string. If C<append> is not 0, the
8457 line is appended to the SV instead of overwriting it. C<append> should
8458 be set to the byte offset that the appended string should start at
8459 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8465 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8475 PERL_ARGS_ASSERT_SV_GETS;
8477 if (SvTHINKFIRST(sv))
8478 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8479 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8481 However, perlbench says it's slower, because the existing swipe code
8482 is faster than copy on write.
8483 Swings and roundabouts. */
8484 SvUPGRADE(sv, SVt_PV);
8487 /* line is going to be appended to the existing buffer in the sv */
8488 if (PerlIO_isutf8(fp)) {
8490 sv_utf8_upgrade_nomg(sv);
8491 sv_pos_u2b(sv,&append,0);
8493 } else if (SvUTF8(sv)) {
8494 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8500 /* not appending - "clear" the string by setting SvCUR to 0,
8501 * the pv is still avaiable. */
8504 if (PerlIO_isutf8(fp))
8507 if (IN_PERL_COMPILETIME) {
8508 /* we always read code in line mode */
8512 else if (RsSNARF(PL_rs)) {
8513 /* If it is a regular disk file use size from stat() as estimate
8514 of amount we are going to read -- may result in mallocing
8515 more memory than we really need if the layers below reduce
8516 the size we read (e.g. CRLF or a gzip layer).
8519 int fd = PerlIO_fileno(fp);
8520 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8521 const Off_t offset = PerlIO_tell(fp);
8522 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8523 #ifdef PERL_COPY_ON_WRITE
8524 /* Add an extra byte for the sake of copy-on-write's
8525 * buffer reference count. */
8526 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8528 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8535 else if (RsRECORD(PL_rs)) {
8536 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8538 else if (RsPARA(PL_rs)) {
8544 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8545 if (PerlIO_isutf8(fp)) {
8546 rsptr = SvPVutf8(PL_rs, rslen);
8549 if (SvUTF8(PL_rs)) {
8550 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8551 Perl_croak(aTHX_ "Wide character in $/");
8554 /* extract the raw pointer to the record separator */
8555 rsptr = SvPV_const(PL_rs, rslen);
8559 /* rslast is the last character in the record separator
8560 * note we don't use rslast except when rslen is true, so the
8561 * null assign is a placeholder. */
8562 rslast = rslen ? rsptr[rslen - 1] : '\0';
8564 if (rspara) { /* have to do this both before and after */
8565 /* to make sure file boundaries work right */
8569 i = PerlIO_getc(fp);
8573 PerlIO_ungetc(fp,i);
8579 /* See if we know enough about I/O mechanism to cheat it ! */
8581 /* This used to be #ifdef test - it is made run-time test for ease
8582 of abstracting out stdio interface. One call should be cheap
8583 enough here - and may even be a macro allowing compile
8587 if (PerlIO_fast_gets(fp)) {
8589 * We can do buffer based IO operations on this filehandle.
8591 * This means we can bypass a lot of subcalls and process
8592 * the buffer directly, it also means we know the upper bound
8593 * on the amount of data we might read of the current buffer
8594 * into our sv. Knowing this allows us to preallocate the pv
8595 * to be able to hold that maximum, which allows us to simplify
8596 * a lot of logic. */
8599 * We're going to steal some values from the stdio struct
8600 * and put EVERYTHING in the innermost loop into registers.
8602 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8603 STRLEN bpx; /* length of the data in the target sv
8604 used to fix pointers after a SvGROW */
8605 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8606 of data left in the read-ahead buffer.
8607 If 0 then the pv buffer can hold the full
8608 amount left, otherwise this is the amount it
8611 /* Here is some breathtakingly efficient cheating */
8613 /* When you read the following logic resist the urge to think
8614 * of record separators that are 1 byte long. They are an
8615 * uninteresting special (simple) case.
8617 * Instead think of record separators which are at least 2 bytes
8618 * long, and keep in mind that we need to deal with such
8619 * separators when they cross a read-ahead buffer boundary.
8621 * Also consider that we need to gracefully deal with separators
8622 * that may be longer than a single read ahead buffer.
8624 * Lastly do not forget we want to copy the delimiter as well. We
8625 * are copying all data in the file _up_to_and_including_ the separator
8628 * Now that you have all that in mind here is what is happening below:
8630 * 1. When we first enter the loop we do some memory book keeping to see
8631 * how much free space there is in the target SV. (This sub assumes that
8632 * it is operating on the same SV most of the time via $_ and that it is
8633 * going to be able to reuse the same pv buffer each call.) If there is
8634 * "enough" room then we set "shortbuffered" to how much space there is
8635 * and start reading forward.
8637 * 2. When we scan forward we copy from the read-ahead buffer to the target
8638 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8639 * and the end of the of pv, as well as for the "rslast", which is the last
8640 * char of the separator.
8642 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8643 * (which has a "complete" record up to the point we saw rslast) and check
8644 * it to see if it matches the separator. If it does we are done. If it doesn't
8645 * we continue on with the scan/copy.
8647 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8648 * the IO system to read the next buffer. We do this by doing a getc(), which
8649 * returns a single char read (or EOF), and prefills the buffer, and also
8650 * allows us to find out how full the buffer is. We use this information to
8651 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8652 * the returned single char into the target sv, and then go back into scan
8655 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8656 * remaining space in the read-buffer.
8658 * Note that this code despite its twisty-turny nature is pretty darn slick.
8659 * It manages single byte separators, multi-byte cross boundary separators,
8660 * and cross-read-buffer separators cleanly and efficiently at the cost
8661 * of potentially greatly overallocating the target SV.
8667 /* get the number of bytes remaining in the read-ahead buffer
8668 * on first call on a given fp this will return 0.*/
8669 cnt = PerlIO_get_cnt(fp);
8671 /* make sure we have the room */
8672 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8673 /* Not room for all of it
8674 if we are looking for a separator and room for some
8676 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8677 /* just process what we have room for */
8678 shortbuffered = cnt - SvLEN(sv) + append + 1;
8679 cnt -= shortbuffered;
8682 /* ensure that the target sv has enough room to hold
8683 * the rest of the read-ahead buffer */
8685 /* remember that cnt can be negative */
8686 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8690 /* we have enough room to hold the full buffer, lets scream */
8694 /* extract the pointer to sv's string buffer, offset by append as necessary */
8695 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8696 /* extract the point to the read-ahead buffer */
8697 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8699 /* some trace debug output */
8700 DEBUG_P(PerlIO_printf(Perl_debug_log,
8701 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8702 DEBUG_P(PerlIO_printf(Perl_debug_log,
8703 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8705 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8706 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8710 /* if there is stuff left in the read-ahead buffer */
8712 /* if there is a separator */
8714 /* find next rslast */
8717 /* shortcut common case of blank line */
8719 if ((*bp++ = *ptr++) == rslast)
8720 goto thats_all_folks;
8722 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8724 SSize_t got = p - ptr + 1;
8725 Copy(ptr, bp, got, STDCHAR);
8729 goto thats_all_folks;
8731 Copy(ptr, bp, cnt, STDCHAR);
8737 /* no separator, slurp the full buffer */
8738 Copy(ptr, bp, cnt, char); /* this | eat */
8739 bp += cnt; /* screams | dust */
8740 ptr += cnt; /* louder | sed :-) */
8742 assert (!shortbuffered);
8743 goto cannot_be_shortbuffered;
8747 if (shortbuffered) { /* oh well, must extend */
8748 /* we didnt have enough room to fit the line into the target buffer
8749 * so we must extend the target buffer and keep going */
8750 cnt = shortbuffered;
8752 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8754 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8755 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8756 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8760 cannot_be_shortbuffered:
8761 /* we need to refill the read-ahead buffer if possible */
8763 DEBUG_P(PerlIO_printf(Perl_debug_log,
8764 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8765 PTR2UV(ptr),(IV)cnt));
8766 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8768 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8769 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8770 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8771 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8774 call PerlIO_getc() to let it prefill the lookahead buffer
8776 This used to call 'filbuf' in stdio form, but as that behaves like
8777 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8778 another abstraction.
8780 Note we have to deal with the char in 'i' if we are not at EOF
8782 bpx = bp - (STDCHAR*)SvPVX_const(sv);
8783 /* signals might be called here, possibly modifying sv */
8784 i = PerlIO_getc(fp); /* get more characters */
8785 bp = (STDCHAR*)SvPVX_const(sv) + bpx;
8787 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8788 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8789 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8790 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8792 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8793 cnt = PerlIO_get_cnt(fp);
8794 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8795 DEBUG_P(PerlIO_printf(Perl_debug_log,
8796 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8797 PTR2UV(ptr),(IV)cnt));
8799 if (i == EOF) /* all done for ever? */
8800 goto thats_really_all_folks;
8802 /* make sure we have enough space in the target sv */
8803 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8805 SvGROW(sv, bpx + cnt + 2);
8806 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8808 /* copy of the char we got from getc() */
8809 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8811 /* make sure we deal with the i being the last character of a separator */
8812 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8813 goto thats_all_folks;
8817 /* check if we have actually found the separator - only really applies
8819 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8820 memNE((char*)bp - rslen, rsptr, rslen))
8821 goto screamer; /* go back to the fray */
8822 thats_really_all_folks:
8824 cnt += shortbuffered;
8825 DEBUG_P(PerlIO_printf(Perl_debug_log,
8826 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8827 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8828 DEBUG_P(PerlIO_printf(Perl_debug_log,
8829 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8831 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8832 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8834 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8835 DEBUG_P(PerlIO_printf(Perl_debug_log,
8836 "Screamer: done, len=%ld, string=|%.*s|\n",
8837 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8841 /*The big, slow, and stupid way. */
8842 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8843 STDCHAR *buf = NULL;
8844 Newx(buf, 8192, STDCHAR);
8852 const STDCHAR * const bpe = buf + sizeof(buf);
8854 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8855 ; /* keep reading */
8859 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8860 /* Accommodate broken VAXC compiler, which applies U8 cast to
8861 * both args of ?: operator, causing EOF to change into 255
8864 i = (U8)buf[cnt - 1];
8870 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8872 sv_catpvn_nomg(sv, (char *) buf, cnt);
8874 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8876 if (i != EOF && /* joy */
8878 SvCUR(sv) < rslen ||
8879 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8883 * If we're reading from a TTY and we get a short read,
8884 * indicating that the user hit his EOF character, we need
8885 * to notice it now, because if we try to read from the TTY
8886 * again, the EOF condition will disappear.
8888 * The comparison of cnt to sizeof(buf) is an optimization
8889 * that prevents unnecessary calls to feof().
8893 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8897 #ifdef USE_HEAP_INSTEAD_OF_STACK
8902 if (rspara) { /* have to do this both before and after */
8903 while (i != EOF) { /* to make sure file boundaries work right */
8904 i = PerlIO_getc(fp);
8906 PerlIO_ungetc(fp,i);
8912 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8918 Auto-increment of the value in the SV, doing string to numeric conversion
8919 if necessary. Handles 'get' magic and operator overloading.
8925 Perl_sv_inc(pTHX_ SV *const sv)
8934 =for apidoc sv_inc_nomg
8936 Auto-increment of the value in the SV, doing string to numeric conversion
8937 if necessary. Handles operator overloading. Skips handling 'get' magic.
8943 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8950 if (SvTHINKFIRST(sv)) {
8951 if (SvREADONLY(sv)) {
8952 Perl_croak_no_modify();
8956 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8958 i = PTR2IV(SvRV(sv));
8962 else sv_force_normal_flags(sv, 0);
8964 flags = SvFLAGS(sv);
8965 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8966 /* It's (privately or publicly) a float, but not tested as an
8967 integer, so test it to see. */
8969 flags = SvFLAGS(sv);
8971 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8972 /* It's publicly an integer, or privately an integer-not-float */
8973 #ifdef PERL_PRESERVE_IVUV
8977 if (SvUVX(sv) == UV_MAX)
8978 sv_setnv(sv, UV_MAX_P1);
8980 (void)SvIOK_only_UV(sv);
8981 SvUV_set(sv, SvUVX(sv) + 1);
8983 if (SvIVX(sv) == IV_MAX)
8984 sv_setuv(sv, (UV)IV_MAX + 1);
8986 (void)SvIOK_only(sv);
8987 SvIV_set(sv, SvIVX(sv) + 1);
8992 if (flags & SVp_NOK) {
8993 const NV was = SvNVX(sv);
8994 if (LIKELY(!Perl_isinfnan(was)) &&
8995 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8996 was >= NV_OVERFLOWS_INTEGERS_AT) {
8997 /* diag_listed_as: Lost precision when %s %f by 1 */
8998 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8999 "Lost precision when incrementing %" NVff " by 1",
9002 (void)SvNOK_only(sv);
9003 SvNV_set(sv, was + 1.0);
9007 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9008 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9009 Perl_croak_no_modify();
9011 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
9012 if ((flags & SVTYPEMASK) < SVt_PVIV)
9013 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
9014 (void)SvIOK_only(sv);
9019 while (isALPHA(*d)) d++;
9020 while (isDIGIT(*d)) d++;
9021 if (d < SvEND(sv)) {
9022 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
9023 #ifdef PERL_PRESERVE_IVUV
9024 /* Got to punt this as an integer if needs be, but we don't issue
9025 warnings. Probably ought to make the sv_iv_please() that does
9026 the conversion if possible, and silently. */
9027 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9028 /* Need to try really hard to see if it's an integer.
9029 9.22337203685478e+18 is an integer.
9030 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9031 so $a="9.22337203685478e+18"; $a+0; $a++
9032 needs to be the same as $a="9.22337203685478e+18"; $a++
9039 /* sv_2iv *should* have made this an NV */
9040 if (flags & SVp_NOK) {
9041 (void)SvNOK_only(sv);
9042 SvNV_set(sv, SvNVX(sv) + 1.0);
9045 /* I don't think we can get here. Maybe I should assert this
9046 And if we do get here I suspect that sv_setnv will croak. NWC
9048 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9049 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9051 #endif /* PERL_PRESERVE_IVUV */
9052 if (!numtype && ckWARN(WARN_NUMERIC))
9053 not_incrementable(sv);
9054 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9058 while (d >= SvPVX_const(sv)) {
9066 /* MKS: The original code here died if letters weren't consecutive.
9067 * at least it didn't have to worry about non-C locales. The
9068 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9069 * arranged in order (although not consecutively) and that only
9070 * [A-Za-z] are accepted by isALPHA in the C locale.
9072 if (isALPHA_FOLD_NE(*d, 'z')) {
9073 do { ++*d; } while (!isALPHA(*d));
9076 *(d--) -= 'z' - 'a';
9081 *(d--) -= 'z' - 'a' + 1;
9085 /* oh,oh, the number grew */
9086 SvGROW(sv, SvCUR(sv) + 2);
9087 SvCUR_set(sv, SvCUR(sv) + 1);
9088 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9099 Auto-decrement of the value in the SV, doing string to numeric conversion
9100 if necessary. Handles 'get' magic and operator overloading.
9106 Perl_sv_dec(pTHX_ SV *const sv)
9115 =for apidoc sv_dec_nomg
9117 Auto-decrement of the value in the SV, doing string to numeric conversion
9118 if necessary. Handles operator overloading. Skips handling 'get' magic.
9124 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9130 if (SvTHINKFIRST(sv)) {
9131 if (SvREADONLY(sv)) {
9132 Perl_croak_no_modify();
9136 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9138 i = PTR2IV(SvRV(sv));
9142 else sv_force_normal_flags(sv, 0);
9144 /* Unlike sv_inc we don't have to worry about string-never-numbers
9145 and keeping them magic. But we mustn't warn on punting */
9146 flags = SvFLAGS(sv);
9147 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9148 /* It's publicly an integer, or privately an integer-not-float */
9149 #ifdef PERL_PRESERVE_IVUV
9153 if (SvUVX(sv) == 0) {
9154 (void)SvIOK_only(sv);
9158 (void)SvIOK_only_UV(sv);
9159 SvUV_set(sv, SvUVX(sv) - 1);
9162 if (SvIVX(sv) == IV_MIN) {
9163 sv_setnv(sv, (NV)IV_MIN);
9167 (void)SvIOK_only(sv);
9168 SvIV_set(sv, SvIVX(sv) - 1);
9173 if (flags & SVp_NOK) {
9176 const NV was = SvNVX(sv);
9177 if (LIKELY(!Perl_isinfnan(was)) &&
9178 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9179 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9180 /* diag_listed_as: Lost precision when %s %f by 1 */
9181 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9182 "Lost precision when decrementing %" NVff " by 1",
9185 (void)SvNOK_only(sv);
9186 SvNV_set(sv, was - 1.0);
9191 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9192 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9193 Perl_croak_no_modify();
9195 if (!(flags & SVp_POK)) {
9196 if ((flags & SVTYPEMASK) < SVt_PVIV)
9197 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9199 (void)SvIOK_only(sv);
9202 #ifdef PERL_PRESERVE_IVUV
9204 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9205 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9206 /* Need to try really hard to see if it's an integer.
9207 9.22337203685478e+18 is an integer.
9208 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9209 so $a="9.22337203685478e+18"; $a+0; $a--
9210 needs to be the same as $a="9.22337203685478e+18"; $a--
9217 /* sv_2iv *should* have made this an NV */
9218 if (flags & SVp_NOK) {
9219 (void)SvNOK_only(sv);
9220 SvNV_set(sv, SvNVX(sv) - 1.0);
9223 /* I don't think we can get here. Maybe I should assert this
9224 And if we do get here I suspect that sv_setnv will croak. NWC
9226 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9227 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9230 #endif /* PERL_PRESERVE_IVUV */
9231 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9234 /* this define is used to eliminate a chunk of duplicated but shared logic
9235 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9236 * used anywhere but here - yves
9238 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9240 SSize_t ix = ++PL_tmps_ix; \
9241 if (UNLIKELY(ix >= PL_tmps_max)) \
9242 ix = tmps_grow_p(ix); \
9243 PL_tmps_stack[ix] = (AnSv); \
9247 =for apidoc sv_mortalcopy
9249 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9250 The new SV is marked as mortal. It will be destroyed "soon", either by an
9251 explicit call to C<FREETMPS>, or by an implicit call at places such as
9252 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9254 =for apidoc sv_mortalcopy_flags
9256 Like C<sv_mortalcopy>, but the extra C<flags> are passed to the
9262 /* Make a string that will exist for the duration of the expression
9263 * evaluation. Actually, it may have to last longer than that, but
9264 * hopefully we won't free it until it has been assigned to a
9265 * permanent location. */
9268 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9272 if (flags & SV_GMAGIC)
9273 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9275 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9276 PUSH_EXTEND_MORTAL__SV_C(sv);
9282 =for apidoc sv_newmortal
9284 Creates a new null SV which is mortal. The reference count of the SV is
9285 set to 1. It will be destroyed "soon", either by an explicit call to
9286 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9287 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9293 Perl_sv_newmortal(pTHX)
9298 SvFLAGS(sv) = SVs_TEMP;
9299 PUSH_EXTEND_MORTAL__SV_C(sv);
9305 =for apidoc newSVpvn_flags
9307 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9308 characters) into it. The reference count for the
9309 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9310 string. You are responsible for ensuring that the source string is at least
9311 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9312 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9313 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9314 returning. If C<SVf_UTF8> is set, C<s>
9315 is considered to be in UTF-8 and the
9316 C<SVf_UTF8> flag will be set on the new SV.
9317 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9319 #define newSVpvn_utf8(s, len, u) \
9320 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9322 =for apidoc Amnh||SVf_UTF8
9323 =for apidoc Amnh||SVs_TEMP
9329 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9333 /* All the flags we don't support must be zero.
9334 And we're new code so I'm going to assert this from the start. */
9335 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9337 sv_setpvn(sv,s,len);
9339 /* This code used to do a sv_2mortal(), however we now unroll the call to
9340 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9341 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9342 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9343 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9344 * means that we eliminate quite a few steps than it looks - Yves
9345 * (explaining patch by gfx) */
9347 SvFLAGS(sv) |= flags;
9349 if(flags & SVs_TEMP){
9350 PUSH_EXTEND_MORTAL__SV_C(sv);
9357 =for apidoc sv_2mortal
9359 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9360 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9361 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9362 string buffer can be "stolen" if this SV is copied. See also
9363 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9369 Perl_sv_2mortal(pTHX_ SV *const sv)
9376 PUSH_EXTEND_MORTAL__SV_C(sv);
9384 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9385 characters) into it. The reference count for the
9386 SV is set to 1. If C<len> is zero, Perl will compute the length using
9387 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9388 C<NUL> characters and has to have a terminating C<NUL> byte).
9390 This function can cause reliability issues if you are likely to pass in
9391 empty strings that are not null terminated, because it will run
9392 strlen on the string and potentially run past valid memory.
9394 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9395 For string literals use L</newSVpvs> instead. This function will work fine for
9396 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9397 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9403 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9408 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9413 =for apidoc newSVpvn
9415 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9416 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9417 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9418 are responsible for ensuring that the source buffer is at least
9419 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9426 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9430 sv_setpvn(sv,buffer,len);
9435 =for apidoc newSVhek
9437 Creates a new SV from the hash key structure. It will generate scalars that
9438 point to the shared string table where possible. Returns a new (undefined)
9439 SV if C<hek> is NULL.
9445 Perl_newSVhek(pTHX_ const HEK *const hek)
9454 if (HEK_LEN(hek) == HEf_SVKEY) {
9455 return newSVsv(*(SV**)HEK_KEY(hek));
9457 const int flags = HEK_FLAGS(hek);
9458 if (flags & HVhek_WASUTF8) {
9460 Andreas would like keys he put in as utf8 to come back as utf8
9462 STRLEN utf8_len = HEK_LEN(hek);
9463 SV * const sv = newSV_type(SVt_PV);
9464 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9465 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9466 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9469 } else if (flags & HVhek_UNSHARED) {
9470 /* A hash that isn't using shared hash keys has to have
9471 the flag in every key so that we know not to try to call
9472 share_hek_hek on it. */
9474 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9479 /* This will be overwhelminly the most common case. */
9481 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9482 more efficient than sharepvn(). */
9486 sv_upgrade(sv, SVt_PV);
9487 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9488 SvCUR_set(sv, HEK_LEN(hek));
9500 =for apidoc newSVpvn_share
9502 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9503 table. If the string does not already exist in the table, it is
9504 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9505 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9506 is non-zero, that value is used; otherwise the hash is computed.
9507 The string's hash can later be retrieved from the SV
9508 with the C<SvSHARED_HASH()> macro. The idea here is
9509 that as the string table is used for shared hash keys these strings will have
9510 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9516 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9520 bool is_utf8 = FALSE;
9521 const char *const orig_src = src;
9524 STRLEN tmplen = -len;
9526 /* See the note in hv.c:hv_fetch() --jhi */
9527 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9531 PERL_HASH(hash, src, len);
9533 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9534 changes here, update it there too. */
9535 sv_upgrade(sv, SVt_PV);
9536 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9543 if (src != orig_src)
9549 =for apidoc newSVpv_share
9551 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9558 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9560 return newSVpvn_share(src, strlen(src), hash);
9563 #if defined(PERL_IMPLICIT_CONTEXT)
9565 /* pTHX_ magic can't cope with varargs, so this is a no-context
9566 * version of the main function, (which may itself be aliased to us).
9567 * Don't access this version directly.
9571 Perl_newSVpvf_nocontext(const char *const pat, ...)
9577 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9579 va_start(args, pat);
9580 sv = vnewSVpvf(pat, &args);
9587 =for apidoc newSVpvf
9589 Creates a new SV and initializes it with the string formatted like
9596 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9601 PERL_ARGS_ASSERT_NEWSVPVF;
9603 va_start(args, pat);
9604 sv = vnewSVpvf(pat, &args);
9609 /* backend for newSVpvf() and newSVpvf_nocontext() */
9612 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9616 PERL_ARGS_ASSERT_VNEWSVPVF;
9619 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9626 Creates a new SV and copies a floating point value into it.
9627 The reference count for the SV is set to 1.
9633 Perl_newSVnv(pTHX_ const NV n)
9645 Creates a new SV and copies an integer into it. The reference count for the
9652 Perl_newSViv(pTHX_ const IV i)
9658 /* Inlining ONLY the small relevant subset of sv_setiv here
9659 * for performance. Makes a significant difference. */
9661 /* We're starting from SVt_FIRST, so provided that's
9662 * actual 0, we don't have to unset any SV type flags
9663 * to promote to SVt_IV. */
9664 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9666 SET_SVANY_FOR_BODYLESS_IV(sv);
9667 SvFLAGS(sv) |= SVt_IV;
9679 Creates a new SV and copies an unsigned integer into it.
9680 The reference count for the SV is set to 1.
9686 Perl_newSVuv(pTHX_ const UV u)
9690 /* Inlining ONLY the small relevant subset of sv_setuv here
9691 * for performance. Makes a significant difference. */
9693 /* Using ivs is more efficient than using uvs - see sv_setuv */
9694 if (u <= (UV)IV_MAX) {
9695 return newSViv((IV)u);
9700 /* We're starting from SVt_FIRST, so provided that's
9701 * actual 0, we don't have to unset any SV type flags
9702 * to promote to SVt_IV. */
9703 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9705 SET_SVANY_FOR_BODYLESS_IV(sv);
9706 SvFLAGS(sv) |= SVt_IV;
9708 (void)SvIsUV_on(sv);
9717 =for apidoc newSV_type
9719 Creates a new SV, of the type specified. The reference count for the new SV
9726 Perl_newSV_type(pTHX_ const svtype type)
9731 ASSUME(SvTYPE(sv) == SVt_FIRST);
9732 if(type != SVt_FIRST)
9733 sv_upgrade(sv, type);
9738 =for apidoc newRV_noinc
9740 Creates an RV wrapper for an SV. The reference count for the original
9741 SV is B<not> incremented.
9747 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9751 PERL_ARGS_ASSERT_NEWRV_NOINC;
9755 /* We're starting from SVt_FIRST, so provided that's
9756 * actual 0, we don't have to unset any SV type flags
9757 * to promote to SVt_IV. */
9758 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9760 SET_SVANY_FOR_BODYLESS_IV(sv);
9761 SvFLAGS(sv) |= SVt_IV;
9766 SvRV_set(sv, tmpRef);
9771 /* newRV_inc is the official function name to use now.
9772 * newRV_inc is in fact #defined to newRV in sv.h
9776 Perl_newRV(pTHX_ SV *const sv)
9778 PERL_ARGS_ASSERT_NEWRV;
9780 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9786 Creates a new SV which is an exact duplicate of the original SV.
9789 =for apidoc newSVsv_nomg
9791 Like C<newSVsv> but does not process get magic.
9797 Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags)
9803 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9804 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9807 /* Do this here, otherwise we leak the new SV if this croaks. */
9808 if (flags & SV_GMAGIC)
9811 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC);
9816 =for apidoc sv_reset
9818 Underlying implementation for the C<reset> Perl function.
9819 Note that the perl-level function is vaguely deprecated.
9825 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9827 PERL_ARGS_ASSERT_SV_RESET;
9829 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9833 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9835 char todo[PERL_UCHAR_MAX+1];
9838 if (!stash || SvTYPE(stash) != SVt_PVHV)
9841 if (!s) { /* reset ?? searches */
9842 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9844 const U32 count = mg->mg_len / sizeof(PMOP**);
9845 PMOP **pmp = (PMOP**) mg->mg_ptr;
9846 PMOP *const *const end = pmp + count;
9850 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9852 (*pmp)->op_pmflags &= ~PMf_USED;
9860 /* reset variables */
9862 if (!HvARRAY(stash))
9865 Zero(todo, 256, char);
9869 I32 i = (unsigned char)*s;
9873 max = (unsigned char)*s++;
9874 for ( ; i <= max; i++) {
9877 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9879 for (entry = HvARRAY(stash)[i];
9881 entry = HeNEXT(entry))
9886 if (!todo[(U8)*HeKEY(entry)])
9888 gv = MUTABLE_GV(HeVAL(entry));
9892 if (sv && !SvREADONLY(sv)) {
9893 SV_CHECK_THINKFIRST_COW_DROP(sv);
9894 if (!isGV(sv)) SvOK_off(sv);
9899 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9910 Using various gambits, try to get an IO from an SV: the IO slot if its a
9911 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9912 named after the PV if we're a string.
9914 'Get' magic is ignored on the C<sv> passed in, but will be called on
9915 C<SvRV(sv)> if C<sv> is an RV.
9921 Perl_sv_2io(pTHX_ SV *const sv)
9926 PERL_ARGS_ASSERT_SV_2IO;
9928 switch (SvTYPE(sv)) {
9930 io = MUTABLE_IO(sv);
9934 if (isGV_with_GP(sv)) {
9935 gv = MUTABLE_GV(sv);
9938 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9939 HEKfARG(GvNAME_HEK(gv)));
9945 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9947 SvGETMAGIC(SvRV(sv));
9948 return sv_2io(SvRV(sv));
9950 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9957 if (SvGMAGICAL(sv)) {
9958 newsv = sv_newmortal();
9959 sv_setsv_nomg(newsv, sv);
9961 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9971 Using various gambits, try to get a CV from an SV; in addition, try if
9972 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9973 The flags in C<lref> are passed to C<gv_fetchsv>.
9979 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9984 PERL_ARGS_ASSERT_SV_2CV;
9991 switch (SvTYPE(sv)) {
9995 return MUTABLE_CV(sv);
10005 sv = amagic_deref_call(sv, to_cv_amg);
10008 if (SvTYPE(sv) == SVt_PVCV) {
10009 cv = MUTABLE_CV(sv);
10014 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
10015 gv = MUTABLE_GV(sv);
10017 Perl_croak(aTHX_ "Not a subroutine reference");
10019 else if (isGV_with_GP(sv)) {
10020 gv = MUTABLE_GV(sv);
10023 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
10030 /* Some flags to gv_fetchsv mean don't really create the GV */
10031 if (!isGV_with_GP(gv)) {
10035 *st = GvESTASH(gv);
10036 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10037 /* XXX this is probably not what they think they're getting.
10038 * It has the same effect as "sub name;", i.e. just a forward
10047 =for apidoc sv_true
10049 Returns true if the SV has a true value by Perl's rules.
10050 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10051 instead use an in-line version.
10057 Perl_sv_true(pTHX_ SV *const sv)
10062 const XPV* const tXpv = (XPV*)SvANY(sv);
10064 (tXpv->xpv_cur > 1 ||
10065 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10072 return SvIVX(sv) != 0;
10075 return SvNVX(sv) != 0.0;
10077 return sv_2bool(sv);
10083 =for apidoc sv_pvn_force
10085 Get a sensible string out of the SV somehow.
10086 A private implementation of the C<SvPV_force> macro for compilers which
10087 can't cope with complex macro expressions. Always use the macro instead.
10089 =for apidoc sv_pvn_force_flags
10091 Get a sensible string out of the SV somehow.
10092 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10093 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10094 implemented in terms of this function.
10095 You normally want to use the various wrapper macros instead: see
10096 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10102 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10104 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10106 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10107 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10108 sv_force_normal_flags(sv, 0);
10118 if (SvTYPE(sv) > SVt_PVLV
10119 || isGV_with_GP(sv))
10120 /* diag_listed_as: Can't coerce %s to %s in %s */
10121 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10123 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10130 if (SvTYPE(sv) < SVt_PV ||
10131 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10134 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10135 SvGROW(sv, len + 1);
10136 Move(s,SvPVX(sv),len,char);
10137 SvCUR_set(sv, len);
10138 SvPVX(sv)[len] = '\0';
10141 SvPOK_on(sv); /* validate pointer */
10143 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10144 PTR2UV(sv),SvPVX_const(sv)));
10147 (void)SvPOK_only_UTF8(sv);
10148 return SvPVX_mutable(sv);
10152 =for apidoc sv_pvbyten_force
10154 The backend for the C<SvPVbytex_force> macro. Always use the macro
10161 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10163 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10165 sv_pvn_force(sv,lp);
10166 sv_utf8_downgrade(sv,0);
10172 =for apidoc sv_pvutf8n_force
10174 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10181 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10183 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10185 sv_pvn_force(sv,0);
10186 sv_utf8_upgrade_nomg(sv);
10192 =for apidoc sv_reftype
10194 Returns a string describing what the SV is a reference to.
10196 If ob is true and the SV is blessed, the string is the class name,
10197 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10203 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10205 PERL_ARGS_ASSERT_SV_REFTYPE;
10206 if (ob && SvOBJECT(sv)) {
10207 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10210 /* WARNING - There is code, for instance in mg.c, that assumes that
10211 * the only reason that sv_reftype(sv,0) would return a string starting
10212 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10213 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10214 * this routine inside other subs, and it saves time.
10215 * Do not change this assumption without searching for "dodgy type check" in
10218 switch (SvTYPE(sv)) {
10233 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10234 /* tied lvalues should appear to be
10235 * scalars for backwards compatibility */
10236 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10237 ? "SCALAR" : "LVALUE");
10238 case SVt_PVAV: return "ARRAY";
10239 case SVt_PVHV: return "HASH";
10240 case SVt_PVCV: return "CODE";
10241 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10242 ? "GLOB" : "SCALAR");
10243 case SVt_PVFM: return "FORMAT";
10244 case SVt_PVIO: return "IO";
10245 case SVt_INVLIST: return "INVLIST";
10246 case SVt_REGEXP: return "REGEXP";
10247 default: return "UNKNOWN";
10255 Returns a SV describing what the SV passed in is a reference to.
10257 dst can be a SV to be set to the description or NULL, in which case a
10258 mortal SV is returned.
10260 If ob is true and the SV is blessed, the description is the class
10261 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10267 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10269 PERL_ARGS_ASSERT_SV_REF;
10272 dst = sv_newmortal();
10274 if (ob && SvOBJECT(sv)) {
10275 HvNAME_get(SvSTASH(sv))
10276 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10277 : sv_setpvs(dst, "__ANON__");
10280 const char * reftype = sv_reftype(sv, 0);
10281 sv_setpv(dst, reftype);
10287 =for apidoc sv_isobject
10289 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10290 object. If the SV is not an RV, or if the object is not blessed, then this
10297 Perl_sv_isobject(pTHX_ SV *sv)
10313 Returns a boolean indicating whether the SV is blessed into the specified
10314 class. This does not check for subtypes; use C<sv_derived_from> to verify
10315 an inheritance relationship.
10321 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10323 const char *hvname;
10325 PERL_ARGS_ASSERT_SV_ISA;
10335 hvname = HvNAME_get(SvSTASH(sv));
10339 return strEQ(hvname, name);
10343 =for apidoc newSVrv
10345 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10346 RV then it will be upgraded to one. If C<classname> is non-null then the new
10347 SV will be blessed in the specified package. The new SV is returned and its
10348 reference count is 1. The reference count 1 is owned by C<rv>. See also
10349 newRV_inc() and newRV_noinc() for creating a new RV properly.
10355 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10359 PERL_ARGS_ASSERT_NEWSVRV;
10363 SV_CHECK_THINKFIRST_COW_DROP(rv);
10365 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10366 const U32 refcnt = SvREFCNT(rv);
10370 SvREFCNT(rv) = refcnt;
10372 sv_upgrade(rv, SVt_IV);
10373 } else if (SvROK(rv)) {
10374 SvREFCNT_dec(SvRV(rv));
10376 prepare_SV_for_RV(rv);
10384 HV* const stash = gv_stashpv(classname, GV_ADD);
10385 (void)sv_bless(rv, stash);
10391 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10393 SV * const lv = newSV_type(SVt_PVLV);
10394 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10396 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10397 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10398 LvSTARGOFF(lv) = ix;
10399 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10404 =for apidoc sv_setref_pv
10406 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10407 argument will be upgraded to an RV. That RV will be modified to point to
10408 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10409 into the SV. The C<classname> argument indicates the package for the
10410 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10411 will have a reference count of 1, and the RV will be returned.
10413 Do not use with other Perl types such as HV, AV, SV, CV, because those
10414 objects will become corrupted by the pointer copy process.
10416 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10422 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10424 PERL_ARGS_ASSERT_SV_SETREF_PV;
10431 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10436 =for apidoc sv_setref_iv
10438 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10439 argument will be upgraded to an RV. That RV will be modified to point to
10440 the new SV. The C<classname> argument indicates the package for the
10441 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10442 will have a reference count of 1, and the RV will be returned.
10448 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10450 PERL_ARGS_ASSERT_SV_SETREF_IV;
10452 sv_setiv(newSVrv(rv,classname), iv);
10457 =for apidoc sv_setref_uv
10459 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10460 argument will be upgraded to an RV. That RV will be modified to point to
10461 the new SV. The C<classname> argument indicates the package for the
10462 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10463 will have a reference count of 1, and the RV will be returned.
10469 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10471 PERL_ARGS_ASSERT_SV_SETREF_UV;
10473 sv_setuv(newSVrv(rv,classname), uv);
10478 =for apidoc sv_setref_nv
10480 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10481 argument will be upgraded to an RV. That RV will be modified to point to
10482 the new SV. The C<classname> argument indicates the package for the
10483 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10484 will have a reference count of 1, and the RV will be returned.
10490 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10492 PERL_ARGS_ASSERT_SV_SETREF_NV;
10494 sv_setnv(newSVrv(rv,classname), nv);
10499 =for apidoc sv_setref_pvn
10501 Copies a string into a new SV, optionally blessing the SV. The length of the
10502 string must be specified with C<n>. The C<rv> argument will be upgraded to
10503 an RV. That RV will be modified to point to the new SV. The C<classname>
10504 argument indicates the package for the blessing. Set C<classname> to
10505 C<NULL> to avoid the blessing. The new SV will have a reference count
10506 of 1, and the RV will be returned.
10508 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10514 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10515 const char *const pv, const STRLEN n)
10517 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10519 sv_setpvn(newSVrv(rv,classname), pv, n);
10524 =for apidoc sv_bless
10526 Blesses an SV into a specified package. The SV must be an RV. The package
10527 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10528 of the SV is unaffected.
10534 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10537 HV *oldstash = NULL;
10539 PERL_ARGS_ASSERT_SV_BLESS;
10543 Perl_croak(aTHX_ "Can't bless non-reference value");
10545 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10546 if (SvREADONLY(tmpRef))
10547 Perl_croak_no_modify();
10548 if (SvOBJECT(tmpRef)) {
10549 oldstash = SvSTASH(tmpRef);
10552 SvOBJECT_on(tmpRef);
10553 SvUPGRADE(tmpRef, SVt_PVMG);
10554 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10555 SvREFCNT_dec(oldstash);
10557 if(SvSMAGICAL(tmpRef))
10558 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10566 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10567 * as it is after unglobbing it.
10570 PERL_STATIC_INLINE void
10571 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10575 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10577 PERL_ARGS_ASSERT_SV_UNGLOB;
10579 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10581 if (!(flags & SV_COW_DROP_PV))
10582 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10584 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10586 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10587 && HvNAME_get(stash))
10588 mro_method_changed_in(stash);
10589 gp_free(MUTABLE_GV(sv));
10592 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10593 GvSTASH(sv) = NULL;
10596 if (GvNAME_HEK(sv)) {
10597 unshare_hek(GvNAME_HEK(sv));
10599 isGV_with_GP_off(sv);
10601 if(SvTYPE(sv) == SVt_PVGV) {
10602 /* need to keep SvANY(sv) in the right arena */
10603 xpvmg = new_XPVMG();
10604 StructCopy(SvANY(sv), xpvmg, XPVMG);
10605 del_XPVGV(SvANY(sv));
10608 SvFLAGS(sv) &= ~SVTYPEMASK;
10609 SvFLAGS(sv) |= SVt_PVMG;
10612 /* Intentionally not calling any local SET magic, as this isn't so much a
10613 set operation as merely an internal storage change. */
10614 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10615 else sv_setsv_flags(sv, temp, 0);
10617 if ((const GV *)sv == PL_last_in_gv)
10618 PL_last_in_gv = NULL;
10619 else if ((const GV *)sv == PL_statgv)
10624 =for apidoc sv_unref_flags
10626 Unsets the RV status of the SV, and decrements the reference count of
10627 whatever was being referenced by the RV. This can almost be thought of
10628 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10629 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10630 (otherwise the decrementing is conditional on the reference count being
10631 different from one or the reference being a readonly SV).
10632 See C<L</SvROK_off>>.
10634 =for apidoc Amnh||SV_IMMEDIATE_UNREF
10640 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10642 SV* const target = SvRV(ref);
10644 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10646 if (SvWEAKREF(ref)) {
10647 sv_del_backref(target, ref);
10648 SvWEAKREF_off(ref);
10649 SvRV_set(ref, NULL);
10652 SvRV_set(ref, NULL);
10654 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10655 assigned to as BEGIN {$a = \"Foo"} will fail. */
10656 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10657 SvREFCNT_dec_NN(target);
10658 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10659 sv_2mortal(target); /* Schedule for freeing later */
10663 =for apidoc sv_untaint
10665 Untaint an SV. Use C<SvTAINTED_off> instead.
10671 Perl_sv_untaint(pTHX_ SV *const sv)
10673 PERL_ARGS_ASSERT_SV_UNTAINT;
10674 PERL_UNUSED_CONTEXT;
10676 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10677 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10684 =for apidoc sv_tainted
10686 Test an SV for taintedness. Use C<SvTAINTED> instead.
10692 Perl_sv_tainted(pTHX_ SV *const sv)
10694 PERL_ARGS_ASSERT_SV_TAINTED;
10695 PERL_UNUSED_CONTEXT;
10697 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10698 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10699 if (mg && (mg->mg_len & 1) )
10705 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10706 private to this file */
10709 =for apidoc sv_setpviv
10711 Copies an integer into the given SV, also updating its string value.
10712 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10718 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10720 /* The purpose of this union is to ensure that arr is aligned on
10721 a 2 byte boundary, because that is what uiv_2buf() requires */
10723 char arr[TYPE_CHARS(UV)];
10727 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10729 PERL_ARGS_ASSERT_SV_SETPVIV;
10731 sv_setpvn(sv, ptr, ebuf - ptr);
10735 =for apidoc sv_setpviv_mg
10737 Like C<sv_setpviv>, but also handles 'set' magic.
10743 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10745 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10747 GCC_DIAG_IGNORE_STMT(-Wdeprecated-declarations);
10749 sv_setpviv(sv, iv);
10751 GCC_DIAG_RESTORE_STMT;
10756 #endif /* NO_MATHOMS */
10758 #if defined(PERL_IMPLICIT_CONTEXT)
10760 /* pTHX_ magic can't cope with varargs, so this is a no-context
10761 * version of the main function, (which may itself be aliased to us).
10762 * Don't access this version directly.
10766 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10771 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10773 va_start(args, pat);
10774 sv_vsetpvf(sv, pat, &args);
10778 /* pTHX_ magic can't cope with varargs, so this is a no-context
10779 * version of the main function, (which may itself be aliased to us).
10780 * Don't access this version directly.
10784 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10789 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10791 va_start(args, pat);
10792 sv_vsetpvf_mg(sv, pat, &args);
10798 =for apidoc sv_setpvf
10800 Works like C<sv_catpvf> but copies the text into the SV instead of
10801 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10807 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10811 PERL_ARGS_ASSERT_SV_SETPVF;
10813 va_start(args, pat);
10814 sv_vsetpvf(sv, pat, &args);
10819 =for apidoc sv_vsetpvf
10821 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10822 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10824 Usually used via its frontend C<sv_setpvf>.
10830 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10832 PERL_ARGS_ASSERT_SV_VSETPVF;
10834 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10838 =for apidoc sv_setpvf_mg
10840 Like C<sv_setpvf>, but also handles 'set' magic.
10846 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10850 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10852 va_start(args, pat);
10853 sv_vsetpvf_mg(sv, pat, &args);
10858 =for apidoc sv_vsetpvf_mg
10860 Like C<sv_vsetpvf>, but also handles 'set' magic.
10862 Usually used via its frontend C<sv_setpvf_mg>.
10868 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10870 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10872 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10876 #if defined(PERL_IMPLICIT_CONTEXT)
10878 /* pTHX_ magic can't cope with varargs, so this is a no-context
10879 * version of the main function, (which may itself be aliased to us).
10880 * Don't access this version directly.
10884 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10889 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10891 va_start(args, pat);
10892 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10896 /* pTHX_ magic can't cope with varargs, so this is a no-context
10897 * version of the main function, (which may itself be aliased to us).
10898 * Don't access this version directly.
10902 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10907 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10909 va_start(args, pat);
10910 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10917 =for apidoc sv_catpvf
10919 Processes its arguments like C<sprintf>, and appends the formatted
10920 output to an SV. As with C<sv_vcatpvfn> called with a non-null C-style
10921 variable argument list, argument reordering is not supported.
10922 If the appended data contains "wide" characters
10923 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10924 and characters >255 formatted with C<%c>), the original SV might get
10925 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10926 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10927 valid UTF-8; if the original SV was bytes, the pattern should be too.
10932 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10936 PERL_ARGS_ASSERT_SV_CATPVF;
10938 va_start(args, pat);
10939 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10944 =for apidoc sv_vcatpvf
10946 Processes its arguments like C<sv_vcatpvfn> called with a non-null C-style
10947 variable argument list, and appends the formatted output
10948 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10950 Usually used via its frontend C<sv_catpvf>.
10956 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10958 PERL_ARGS_ASSERT_SV_VCATPVF;
10960 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10964 =for apidoc sv_catpvf_mg
10966 Like C<sv_catpvf>, but also handles 'set' magic.
10972 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10976 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10978 va_start(args, pat);
10979 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10985 =for apidoc sv_vcatpvf_mg
10987 Like C<sv_vcatpvf>, but also handles 'set' magic.
10989 Usually used via its frontend C<sv_catpvf_mg>.
10995 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10997 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10999 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
11004 =for apidoc sv_vsetpvfn
11006 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
11009 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
11015 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11016 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11018 PERL_ARGS_ASSERT_SV_VSETPVFN;
11021 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
11025 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
11027 PERL_STATIC_INLINE void
11028 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
11030 STRLEN const need = len + SvCUR(sv) + 1;
11033 /* can't wrap as both len and SvCUR() are allocated in
11034 * memory and together can't consume all the address space
11036 assert(need > len);
11041 Copy(buf, end, len, char);
11044 SvCUR_set(sv, need - 1);
11049 * Warn of missing argument to sprintf. The value used in place of such
11050 * arguments should be &PL_sv_no; an undefined value would yield
11051 * inappropriate "use of uninit" warnings [perl #71000].
11054 S_warn_vcatpvfn_missing_argument(pTHX) {
11055 if (ckWARN(WARN_MISSING)) {
11056 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11057 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11066 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11067 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11071 /* Given an int i from the next arg (if args is true) or an sv from an arg
11072 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11073 * with overflow checking.
11074 * Sets *neg to true if the value was negative (untouched otherwise.
11075 * Returns the absolute value.
11076 * As an extra margin of safety, it croaks if the returned value would
11077 * exceed the maximum value of a STRLEN / 4.
11081 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11095 if (UNLIKELY(SvIsUV(sv))) {
11096 UV uv = SvUV_nomg(sv);
11098 S_croak_overflow();
11102 iv = SvIV_nomg(sv);
11106 S_croak_overflow();
11112 if (iv > (IV)(((STRLEN)~0) / 4))
11113 S_croak_overflow();
11118 /* Read in and return a number. Updates *pattern to point to the char
11119 * following the number. Expects the first char to 1..9.
11120 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11121 * This is a belt-and-braces safety measure to complement any
11122 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11123 * It means that e.g. on a 32-bit system the width/precision can't be more
11124 * than 1G, which seems reasonable.
11128 S_expect_number(pTHX_ const char **const pattern)
11132 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11134 assert(inRANGE(**pattern, '1', '9'));
11136 var = *(*pattern)++ - '0';
11137 while (isDIGIT(**pattern)) {
11138 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11139 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11140 S_croak_overflow();
11141 var = var * 10 + (*(*pattern)++ - '0');
11146 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11147 * ensures it's big enough), back fill it with the rounded integer part of
11148 * nv. Returns ptr to start of string, and sets *len to its length.
11149 * Returns NULL if not convertible.
11153 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11155 const int neg = nv < 0;
11158 PERL_ARGS_ASSERT_F0CONVERT;
11160 assert(!Perl_isinfnan(nv));
11163 if (nv != 0.0 && nv < UV_MAX) {
11169 if (uv & 1 && uv == nv)
11170 uv--; /* Round to even */
11173 const unsigned dig = uv % 10;
11175 } while (uv /= 10);
11185 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11188 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11189 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11191 PERL_ARGS_ASSERT_SV_VCATPVFN;
11193 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11197 /* For the vcatpvfn code, we need a long double target in case
11198 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11199 * with long double formats, even without NV being long double. But we
11200 * call the target 'fv' instead of 'nv', since most of the time it is not
11201 * (most compilers these days recognize "long double", even if only as a
11202 * synonym for "double").
11204 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11205 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11206 # define VCATPVFN_FV_GF PERL_PRIgldbl
11207 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11208 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11209 # define VCATPVFN_NV_TO_FV(nv,fv) \
11212 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11215 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11217 typedef long double vcatpvfn_long_double_t;
11219 # define VCATPVFN_FV_GF NVgf
11220 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11221 typedef NV vcatpvfn_long_double_t;
11224 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11225 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11226 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11227 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11228 * after the first 1023 zero bits.
11230 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11231 * of dynamically growing buffer might be better, start at just 16 bytes
11232 * (for example) and grow only when necessary. Or maybe just by looking
11233 * at the exponents of the two doubles? */
11234 # define DOUBLEDOUBLE_MAXBITS 2098
11237 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11238 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11239 * per xdigit. For the double-double case, this can be rather many.
11240 * The non-double-double-long-double overshoots since all bits of NV
11241 * are not mantissa bits, there are also exponent bits. */
11242 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11243 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11245 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11248 /* If we do not have a known long double format, (including not using
11249 * long doubles, or long doubles being equal to doubles) then we will
11250 * fall back to the ldexp/frexp route, with which we can retrieve at
11251 * most as many bits as our widest unsigned integer type is. We try
11252 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11254 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11255 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11257 #if defined(HAS_QUAD) && defined(Uquad_t)
11258 # define MANTISSATYPE Uquad_t
11259 # define MANTISSASIZE 8
11261 # define MANTISSATYPE UV
11262 # define MANTISSASIZE UVSIZE
11265 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11266 # define HEXTRACT_LITTLE_ENDIAN
11267 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11268 # define HEXTRACT_BIG_ENDIAN
11270 # define HEXTRACT_MIX_ENDIAN
11273 /* S_hextract() is a helper for S_format_hexfp, for extracting
11274 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11275 * are being extracted from (either directly from the long double in-memory
11276 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11277 * is used to update the exponent. The subnormal is set to true
11278 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11279 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11281 * The tricky part is that S_hextract() needs to be called twice:
11282 * the first time with vend as NULL, and the second time with vend as
11283 * the pointer returned by the first call. What happens is that on
11284 * the first round the output size is computed, and the intended
11285 * extraction sanity checked. On the second round the actual output
11286 * (the extraction of the hexadecimal values) takes place.
11287 * Sanity failures cause fatal failures during both rounds. */
11289 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11290 U8* vhex, U8* vend)
11294 int ixmin = 0, ixmax = 0;
11296 /* XXX Inf/NaN are not handled here, since it is
11297 * assumed they are to be output as "Inf" and "NaN". */
11299 /* These macros are just to reduce typos, they have multiple
11300 * repetitions below, but usually only one (or sometimes two)
11301 * of them is really being used. */
11302 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11303 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11304 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11305 #define HEXTRACT_OUTPUT(ix) \
11307 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11309 #define HEXTRACT_COUNT(ix, c) \
11311 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11313 #define HEXTRACT_BYTE(ix) \
11315 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11317 #define HEXTRACT_LO_NYBBLE(ix) \
11319 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11321 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11322 * to make it look less odd when the top bits of a NV
11323 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11324 * order bits can be in the "low nybble" of a byte. */
11325 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11326 #define HEXTRACT_BYTES_LE(a, b) \
11327 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11328 #define HEXTRACT_BYTES_BE(a, b) \
11329 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11330 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11331 #define HEXTRACT_IMPLICIT_BIT(nv) \
11333 if (!*subnormal) { \
11334 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11338 /* Most formats do. Those which don't should undef this.
11340 * But also note that IEEE 754 subnormals do not have it, or,
11341 * expressed alternatively, their implicit bit is zero. */
11342 #define HEXTRACT_HAS_IMPLICIT_BIT
11344 /* Many formats do. Those which don't should undef this. */
11345 #define HEXTRACT_HAS_TOP_NYBBLE
11347 /* HEXTRACTSIZE is the maximum number of xdigits. */
11348 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11349 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11351 # define HEXTRACTSIZE 2 * NVSIZE
11354 const U8* vmaxend = vhex + HEXTRACTSIZE;
11356 assert(HEXTRACTSIZE <= VHEX_SIZE);
11358 PERL_UNUSED_VAR(ix); /* might happen */
11359 (void)Perl_frexp(PERL_ABS(nv), exponent);
11360 *subnormal = FALSE;
11361 if (vend && (vend <= vhex || vend > vmaxend)) {
11362 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11363 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11366 /* First check if using long doubles. */
11367 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11368 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11369 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11370 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11371 /* The bytes 13..0 are the mantissa/fraction,
11372 * the 15,14 are the sign+exponent. */
11373 const U8* nvp = (const U8*)(&nv);
11374 HEXTRACT_GET_SUBNORMAL(nv);
11375 HEXTRACT_IMPLICIT_BIT(nv);
11376 # undef HEXTRACT_HAS_TOP_NYBBLE
11377 HEXTRACT_BYTES_LE(13, 0);
11378 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11379 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11380 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11381 /* The bytes 2..15 are the mantissa/fraction,
11382 * the 0,1 are the sign+exponent. */
11383 const U8* nvp = (const U8*)(&nv);
11384 HEXTRACT_GET_SUBNORMAL(nv);
11385 HEXTRACT_IMPLICIT_BIT(nv);
11386 # undef HEXTRACT_HAS_TOP_NYBBLE
11387 HEXTRACT_BYTES_BE(2, 15);
11388 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11389 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11390 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11391 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11392 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11393 /* The bytes 0..1 are the sign+exponent,
11394 * the bytes 2..9 are the mantissa/fraction. */
11395 const U8* nvp = (const U8*)(&nv);
11396 # undef HEXTRACT_HAS_IMPLICIT_BIT
11397 # undef HEXTRACT_HAS_TOP_NYBBLE
11398 HEXTRACT_GET_SUBNORMAL(nv);
11399 HEXTRACT_BYTES_LE(7, 0);
11400 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11401 /* Does this format ever happen? (Wikipedia says the Motorola
11402 * 6888x math coprocessors used format _like_ this but padded
11403 * to 96 bits with 16 unused bits between the exponent and the
11405 const U8* nvp = (const U8*)(&nv);
11406 # undef HEXTRACT_HAS_IMPLICIT_BIT
11407 # undef HEXTRACT_HAS_TOP_NYBBLE
11408 HEXTRACT_GET_SUBNORMAL(nv);
11409 HEXTRACT_BYTES_BE(0, 7);
11411 # define HEXTRACT_FALLBACK
11412 /* Double-double format: two doubles next to each other.
11413 * The first double is the high-order one, exactly like
11414 * it would be for a "lone" double. The second double
11415 * is shifted down using the exponent so that that there
11416 * are no common bits. The tricky part is that the value
11417 * of the double-double is the SUM of the two doubles and
11418 * the second one can be also NEGATIVE.
11420 * Because of this tricky construction the bytewise extraction we
11421 * use for the other long double formats doesn't work, we must
11422 * extract the values bit by bit.
11424 * The little-endian double-double is used .. somewhere?
11426 * The big endian double-double is used in e.g. PPC/Power (AIX)
11429 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11430 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11431 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11434 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11435 /* Using normal doubles, not long doubles.
11437 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11438 * bytes, since we might need to handle printf precision, and
11439 * also need to insert the radix. */
11441 # ifdef HEXTRACT_LITTLE_ENDIAN
11442 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11443 const U8* nvp = (const U8*)(&nv);
11444 HEXTRACT_GET_SUBNORMAL(nv);
11445 HEXTRACT_IMPLICIT_BIT(nv);
11446 HEXTRACT_TOP_NYBBLE(6);
11447 HEXTRACT_BYTES_LE(5, 0);
11448 # elif defined(HEXTRACT_BIG_ENDIAN)
11449 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11450 const U8* nvp = (const U8*)(&nv);
11451 HEXTRACT_GET_SUBNORMAL(nv);
11452 HEXTRACT_IMPLICIT_BIT(nv);
11453 HEXTRACT_TOP_NYBBLE(1);
11454 HEXTRACT_BYTES_BE(2, 7);
11455 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11456 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11457 const U8* nvp = (const U8*)(&nv);
11458 HEXTRACT_GET_SUBNORMAL(nv);
11459 HEXTRACT_IMPLICIT_BIT(nv);
11460 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11461 HEXTRACT_BYTE(1); /* 5 */
11462 HEXTRACT_BYTE(0); /* 4 */
11463 HEXTRACT_BYTE(7); /* 3 */
11464 HEXTRACT_BYTE(6); /* 2 */
11465 HEXTRACT_BYTE(5); /* 1 */
11466 HEXTRACT_BYTE(4); /* 0 */
11467 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11468 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11469 const U8* nvp = (const U8*)(&nv);
11470 HEXTRACT_GET_SUBNORMAL(nv);
11471 HEXTRACT_IMPLICIT_BIT(nv);
11472 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11473 HEXTRACT_BYTE(6); /* 5 */
11474 HEXTRACT_BYTE(7); /* 4 */
11475 HEXTRACT_BYTE(0); /* 3 */
11476 HEXTRACT_BYTE(1); /* 2 */
11477 HEXTRACT_BYTE(2); /* 1 */
11478 HEXTRACT_BYTE(3); /* 0 */
11480 # define HEXTRACT_FALLBACK
11483 # define HEXTRACT_FALLBACK
11485 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11487 #ifdef HEXTRACT_FALLBACK
11488 HEXTRACT_GET_SUBNORMAL(nv);
11489 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11490 /* The fallback is used for the double-double format, and
11491 * for unknown long double formats, and for unknown double
11492 * formats, or in general unknown NV formats. */
11493 if (nv == (NV)0.0) {
11501 NV d = nv < 0 ? -nv : nv;
11503 U8 ha = 0x0; /* hexvalue accumulator */
11504 U8 hd = 0x8; /* hexvalue digit */
11506 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11507 * this is essentially manual frexp(). Multiplying by 0.5 and
11508 * doubling should be lossless in binary floating point. */
11518 while (d >= e + e) {
11522 /* Now e <= d < 2*e */
11524 /* First extract the leading hexdigit (the implicit bit). */
11540 /* Then extract the remaining hexdigits. */
11541 while (d > (NV)0.0) {
11547 /* Output or count in groups of four bits,
11548 * that is, when the hexdigit is down to one. */
11553 /* Reset the hexvalue. */
11562 /* Flush possible pending hexvalue. */
11572 /* Croak for various reasons: if the output pointer escaped the
11573 * output buffer, if the extraction index escaped the extraction
11574 * buffer, or if the ending output pointer didn't match the
11575 * previously computed value. */
11576 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11577 /* For double-double the ixmin and ixmax stay at zero,
11578 * which is convenient since the HEXTRACTSIZE is tricky
11579 * for double-double. */
11580 ixmin < 0 || ixmax >= NVSIZE ||
11581 (vend && v != vend)) {
11582 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11583 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11589 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11591 * Processes the %a/%A hexadecimal floating-point format, since the
11592 * built-in snprintf()s which are used for most of the f/p formats, don't
11593 * universally handle %a/%A.
11594 * Populates buf of length bufsize, and returns the length of the created
11596 * The rest of the args have the same meaning as the local vars of the
11597 * same name within Perl_sv_vcatpvfn_flags().
11599 * The caller's determination of IN_LC(LC_NUMERIC), passed as in_lc_numeric,
11600 * is used to ensure we do the right thing when we need to access the locale's
11603 * It requires the caller to make buf large enough.
11607 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11608 const NV nv, const vcatpvfn_long_double_t fv,
11609 bool has_precis, STRLEN precis, STRLEN width,
11610 bool alt, char plus, bool left, bool fill, bool in_lc_numeric)
11612 /* Hexadecimal floating point. */
11614 U8 vhex[VHEX_SIZE];
11615 U8* v = vhex; /* working pointer to vhex */
11616 U8* vend; /* pointer to one beyond last digit of vhex */
11617 U8* vfnz = NULL; /* first non-zero */
11618 U8* vlnz = NULL; /* last non-zero */
11619 U8* v0 = NULL; /* first output */
11620 const bool lower = (c == 'a');
11621 /* At output the values of vhex (up to vend) will
11622 * be mapped through the xdig to get the actual
11623 * human-readable xdigits. */
11624 const char* xdig = PL_hexdigit;
11625 STRLEN zerotail = 0; /* how many extra zeros to append */
11626 int exponent = 0; /* exponent of the floating point input */
11627 bool hexradix = FALSE; /* should we output the radix */
11628 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11629 bool negative = FALSE;
11632 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11634 * For example with denormals, (assuming the vanilla
11635 * 64-bit double): the exponent is zero. 1xp-1074 is
11636 * the smallest denormal and the smallest double, it
11637 * could be output also as 0x0.0000000000001p-1022 to
11638 * match its internal structure. */
11640 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11641 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11643 #if NVSIZE > DOUBLESIZE
11644 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11645 /* In this case there is an implicit bit,
11646 * and therefore the exponent is shifted by one. */
11648 # elif defined(NV_X86_80_BIT)
11650 /* The subnormals of the x86-80 have a base exponent of -16382,
11651 * (while the physical exponent bits are zero) but the frexp()
11652 * returned the scientific-style floating exponent. We want
11653 * to map the last one as:
11654 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11655 * -16835..-16388 -> -16384
11656 * since we want to keep the first hexdigit
11657 * as one of the [8421]. */
11658 exponent = -4 * ( (exponent + 1) / -4) - 2;
11662 /* TBD: other non-implicit-bit platforms than the x86-80. */
11666 negative = fv < 0 || Perl_signbit(nv);
11677 xdig += 16; /* Use uppercase hex. */
11680 /* Find the first non-zero xdigit. */
11681 for (v = vhex; v < vend; v++) {
11689 /* Find the last non-zero xdigit. */
11690 for (v = vend - 1; v >= vhex; v--) {
11697 #if NVSIZE == DOUBLESIZE
11703 #ifndef NV_X86_80_BIT
11705 /* IEEE 754 subnormals (but not the x86 80-bit):
11706 * we want "normalize" the subnormal,
11707 * so we need to right shift the hex nybbles
11708 * so that the output of the subnormal starts
11709 * from the first true bit. (Another, equally
11710 * valid, policy would be to dump the subnormal
11711 * nybbles as-is, to display the "physical" layout.) */
11714 /* Find the ceil(log2(v[0])) of
11715 * the top non-zero nybble. */
11716 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11720 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11721 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11735 U8* ve = (subnormal ? vlnz + 1 : vend);
11736 SSize_t vn = ve - v0;
11738 if (precis < (Size_t)(vn - 1)) {
11739 bool overflow = FALSE;
11740 if (v0[precis + 1] < 0x8) {
11741 /* Round down, nothing to do. */
11742 } else if (v0[precis + 1] > 0x8) {
11745 overflow = v0[precis] > 0xF;
11747 } else { /* v0[precis] == 0x8 */
11748 /* Half-point: round towards the one
11749 * with the even least-significant digit:
11757 * 78 -> 8 f8 -> 10 */
11758 if ((v0[precis] & 0x1)) {
11761 overflow = v0[precis] > 0xF;
11766 for (v = v0 + precis - 1; v >= v0; v--) {
11768 overflow = *v > 0xF;
11774 if (v == v0 - 1 && overflow) {
11775 /* If the overflow goes all the
11776 * way to the front, we need to
11777 * insert 0x1 in front, and adjust
11779 Move(v0, v0 + 1, vn - 1, char);
11785 /* The new effective "last non zero". */
11786 vlnz = v0 + precis;
11790 subnormal ? precis - vn + 1 :
11791 precis - (vlnz - vhex);
11798 /* If there are non-zero xdigits, the radix
11799 * is output after the first one. */
11807 zerotail = has_precis ? precis : 0;
11810 /* The radix is always output if precis, or if alt. */
11811 if ((has_precis && precis > 0) || alt) {
11816 #ifndef USE_LOCALE_NUMERIC
11819 if (in_lc_numeric) {
11821 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, {
11822 const char* r = SvPV(PL_numeric_radix_sv, n);
11823 Copy(r, p, n, char);
11838 if (zerotail > 0) {
11839 while (zerotail--) {
11846 /* sanity checks */
11847 if (elen >= bufsize || width >= bufsize)
11848 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11849 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11851 elen += my_snprintf(p, bufsize - elen,
11852 "%c%+d", lower ? 'p' : 'P',
11855 if (elen < width) {
11856 STRLEN gap = (STRLEN)(width - elen);
11858 /* Pad the back with spaces. */
11859 memset(buf + elen, ' ', gap);
11862 /* Insert the zeros after the "0x" and the
11863 * the potential sign, but before the digits,
11864 * otherwise we end up with "0000xH.HHH...",
11865 * when we want "0x000H.HHH..." */
11866 STRLEN nzero = gap;
11867 char* zerox = buf + 2;
11868 STRLEN nmove = elen - 2;
11869 if (negative || plus) {
11873 Move(zerox, zerox + nzero, nmove, char);
11874 memset(zerox, fill ? '0' : ' ', nzero);
11877 /* Move it to the right. */
11878 Move(buf, buf + gap,
11880 /* Pad the front with spaces. */
11881 memset(buf, ' ', gap);
11890 =for apidoc sv_vcatpvfn
11892 =for apidoc sv_vcatpvfn_flags
11894 Processes its arguments like C<vsprintf> and appends the formatted output
11895 to an SV. Uses an array of SVs if the C-style variable argument list is
11896 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11897 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11898 C<va_list> argument list with a format string that uses argument reordering
11899 will yield an exception.
11901 When running with taint checks enabled, indicates via
11902 C<maybe_tainted> if results are untrustworthy (often due to the use of
11905 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11907 It assumes that pat has the same utf8-ness as sv. It's the caller's
11908 responsibility to ensure that this is so.
11910 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11917 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11918 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11921 const char *fmtstart; /* character following the current '%' */
11922 const char *q; /* current position within format */
11923 const char *patend;
11926 static const char nullstr[] = "(null)";
11927 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11928 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11929 /* Times 4: a decimal digit takes more than 3 binary digits.
11930 * NV_DIG: mantissa takes that many decimal digits.
11931 * Plus 32: Playing safe. */
11932 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11933 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11934 #ifdef USE_LOCALE_NUMERIC
11935 bool have_in_lc_numeric = FALSE;
11937 /* we never change this unless USE_LOCALE_NUMERIC */
11938 bool in_lc_numeric = FALSE;
11940 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11941 PERL_UNUSED_ARG(maybe_tainted);
11943 if (flags & SV_GMAGIC)
11946 /* no matter what, this is a string now */
11947 (void)SvPV_force_nomg(sv, origlen);
11949 /* the code that scans for flags etc following a % relies on
11950 * a '\0' being present to avoid falling off the end. Ideally that
11951 * should be fixed */
11952 assert(pat[patlen] == '\0');
11955 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11956 * In each case, if there isn't the correct number of args, instead
11957 * fall through to the main code to handle the issuing of any
11961 if (patlen == 0 && (args || sv_count == 0))
11964 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11967 if (patlen == 2 && pat[1] == 's') {
11969 const char * const s = va_arg(*args, char*);
11970 sv_catpv_nomg(sv, s ? s : nullstr);
11973 /* we want get magic on the source but not the target.
11974 * sv_catsv can't do that, though */
11975 SvGETMAGIC(*svargs);
11976 sv_catsv_nomg(sv, *svargs);
11983 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11984 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11985 sv_catsv_nomg(sv, asv);
11989 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11990 /* special-case "%.0f" */
11991 else if ( patlen == 4
11992 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11994 const NV nv = SvNV(*svargs);
11995 if (LIKELY(!Perl_isinfnan(nv))) {
11999 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
12000 sv_catpvn_nomg(sv, p, l);
12005 #endif /* !USE_LONG_DOUBLE */
12009 patend = (char*)pat + patlen;
12010 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
12011 char intsize = 0; /* size qualifier in "%hi..." etc */
12012 bool alt = FALSE; /* has "%#..." */
12013 bool left = FALSE; /* has "%-..." */
12014 bool fill = FALSE; /* has "%0..." */
12015 char plus = 0; /* has "%+..." */
12016 STRLEN width = 0; /* value of "%NNN..." */
12017 bool has_precis = FALSE; /* has "%.NNN..." */
12018 STRLEN precis = 0; /* value of "%.NNN..." */
12019 int base = 0; /* base to print in, e.g. 8 for %o */
12020 UV uv = 0; /* the value to print of int-ish args */
12022 bool vectorize = FALSE; /* has "%v..." */
12023 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
12024 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
12025 STRLEN veclen = 0; /* SvCUR(vec arg) */
12026 const char *dotstr = NULL; /* separator string for %v */
12027 STRLEN dotstrlen; /* length of separator string for %v */
12029 Size_t efix = 0; /* explicit format parameter index */
12030 const Size_t osvix = svix; /* original index in case of bad fmt */
12033 bool is_utf8 = FALSE; /* is this item utf8? */
12034 bool arg_missing = FALSE; /* give "Missing argument" warning */
12035 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
12036 STRLEN esignlen = 0; /* length of e.g. "-0x" */
12037 STRLEN zeros = 0; /* how many '0' to prepend */
12039 const char *eptr = NULL; /* the address of the element string */
12040 STRLEN elen = 0; /* the length of the element string */
12042 char c; /* the actual format ('d', s' etc) */
12045 /* echo everything up to the next format specification */
12046 for (q = fmtstart; q < patend && *q != '%'; ++q)
12049 if (q > fmtstart) {
12050 if (has_utf8 && !pat_utf8) {
12051 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12055 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12057 for (p = fmtstart; p < q; p++)
12058 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12063 for (p = fmtstart; p < q; p++)
12064 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12066 SvCUR_set(sv, need - 1);
12069 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12074 fmtstart = q; /* fmtstart is char following the '%' */
12077 We allow format specification elements in this order:
12078 \d+\$ explicit format parameter index
12080 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12081 0 flag (as above): repeated to allow "v02"
12082 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12083 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12085 [%bcdefginopsuxDFOUX] format (mandatory)
12088 if (inRANGE(*q, '1', '9')) {
12089 width = expect_number(&q);
12092 Perl_croak_nocontext(
12093 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12095 efix = (Size_t)width;
12097 no_redundant_warning = TRUE;
12109 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12136 /* at this point we can expect one of:
12138 * 123 an explicit width
12139 * * width taken from next arg
12140 * *12$ width taken from 12th arg
12143 * But any width specification may be preceded by a v, in one of its
12148 * So an asterisk may be either a width specifier or a vector
12149 * separator arg specifier, and we don't know which initially
12154 STRLEN ix; /* explicit width/vector separator index */
12156 if (inRANGE(*q, '1', '9')) {
12157 ix = expect_number(&q);
12160 Perl_croak_nocontext(
12161 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12162 no_redundant_warning = TRUE;
12171 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12172 * with the default "." */
12177 vecsv = va_arg(*args, SV*);
12179 ix = ix ? ix - 1 : svix++;
12180 vecsv = ix < sv_count ? svargs[ix]
12181 : (arg_missing = TRUE, &PL_sv_no);
12183 dotstr = SvPV_const(vecsv, dotstrlen);
12184 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12185 bad with tied or overloaded values that return UTF8. */
12186 if (DO_UTF8(vecsv))
12188 else if (has_utf8) {
12189 vecsv = sv_mortalcopy(vecsv);
12190 sv_utf8_upgrade(vecsv);
12191 dotstr = SvPV_const(vecsv, dotstrlen);
12198 /* the asterisk specified a width */
12203 i = va_arg(*args, int);
12205 ix = ix ? ix - 1 : svix++;
12206 sv = (ix < sv_count) ? svargs[ix]
12207 : (arg_missing = TRUE, (SV*)NULL);
12209 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12212 else if (*q == 'v') {
12223 /* explicit width? */
12228 if (inRANGE(*q, '1', '9'))
12229 width = expect_number(&q);
12239 STRLEN ix; /* explicit precision index */
12241 if (inRANGE(*q, '1', '9')) {
12242 ix = expect_number(&q);
12245 Perl_croak_nocontext(
12246 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12247 no_redundant_warning = TRUE;
12260 i = va_arg(*args, int);
12262 ix = ix ? ix - 1 : svix++;
12263 sv = (ix < sv_count) ? svargs[ix]
12264 : (arg_missing = TRUE, (SV*)NULL);
12266 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12268 /* ignore negative precision */
12274 /* although it doesn't seem documented, this code has long
12276 * no digits following the '.' is treated like '.0'
12277 * the number may be preceded by any number of zeroes,
12278 * e.g. "%.0001f", which is the same as "%.1f"
12279 * so I've kept that behaviour. DAPM May 2017
12283 precis = inRANGE(*q, '1', '9') ? expect_number(&q) : 0;
12292 case 'I': /* Ix, I32x, and I64x */
12293 # ifdef USE_64_BIT_INT
12294 if (q[1] == '6' && q[2] == '4') {
12300 if (q[1] == '3' && q[2] == '2') {
12304 # ifdef USE_64_BIT_INT
12310 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12311 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12314 # ifdef USE_QUADMATH
12327 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12328 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12329 if (*q == 'l') { /* lld, llf */
12338 if (*++q == 'h') { /* hhd, hhu */
12355 c = *q++; /* c now holds the conversion type */
12357 /* '%' doesn't have an arg, so skip arg processing */
12366 if (vectorize && !strchr("BbDdiOouUXx", c))
12369 /* get next arg (individual branches do their own va_arg()
12370 * handling for the args case) */
12373 efix = efix ? efix - 1 : svix++;
12374 argsv = efix < sv_count ? svargs[efix]
12375 : (arg_missing = TRUE, &PL_sv_no);
12385 eptr = va_arg(*args, char*);
12388 elen = my_strnlen(eptr, precis);
12390 elen = strlen(eptr);
12392 eptr = (char *)nullstr;
12393 elen = sizeof nullstr - 1;
12397 eptr = SvPV_const(argsv, elen);
12398 if (DO_UTF8(argsv)) {
12399 STRLEN old_precis = precis;
12400 if (has_precis && precis < elen) {
12401 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12402 STRLEN p = precis > ulen ? ulen : precis;
12403 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12404 /* sticks at end */
12406 if (width) { /* fudge width (can't fudge elen) */
12407 if (has_precis && precis < elen)
12408 width += precis - old_precis;
12411 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12418 if (has_precis && precis < elen)
12430 * "%...p" is normally treated like "%...x", except that the
12431 * number to print is the SV's address (or a pointer address
12432 * for C-ish sprintf).
12434 * However, the C-ish sprintf variant allows a few special
12435 * extensions. These are currently:
12437 * %-p (SVf) Like %s, but gets the string from an SV*
12438 * arg rather than a char* arg.
12439 * (This was previously %_).
12441 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12443 * %2p (HEKf) Like %s, but using the key string in a HEK
12445 * %3p (HEKf256) Ditto but like %.256s
12447 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12448 * (cBOOL(utf8), len, string_buf).
12449 * It's handled by the "case 'd'" branch
12450 * rather than here.
12452 * %<num>p where num is 1 or > 4: reserved for future
12453 * extensions. Warns, but then is treated as a
12454 * general %p (print hex address) format.
12462 /* not %*p or %*1$p - any width was explicit */
12466 if (left) { /* %-p (SVf), %-NNNp */
12471 argsv = MUTABLE_SV(va_arg(*args, void*));
12472 eptr = SvPV_const(argsv, elen);
12473 if (DO_UTF8(argsv))
12478 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12479 HEK * const hek = va_arg(*args, HEK *);
12480 eptr = HEK_KEY(hek);
12481 elen = HEK_LEN(hek);
12492 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12493 "internal %%<num>p might conflict with future printf extensions");
12497 /* treat as normal %...p */
12499 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12504 /* Ignore any size specifiers, since they're not documented as
12505 * being allowed for %c (ideally we should warn on e.g. '%hc').
12506 * Setting a default intsize, along with a positive
12507 * (which signals unsigned) base, causes, for C-ish use, the
12508 * va_arg to be interpreted as as unsigned int, when it's
12509 * actually signed, which will convert -ve values to high +ve
12510 * values. Note that unlike the libc %c, values > 255 will
12511 * convert to high unicode points rather than being truncated
12512 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12513 * will again convert -ve args to high -ve values.
12516 base = 1; /* special value that indicates we're doing a 'c' */
12517 goto get_int_arg_val;
12526 goto get_int_arg_val;
12529 /* probably just a plain %d, but it might be the start of the
12530 * special UTF8f format, which usually looks something like
12531 * "%d%lu%4p" (the lu may vary by platform)
12533 assert((UTF8f)[0] == 'd');
12534 assert((UTF8f)[1] == '%');
12536 if ( args /* UTF8f only valid for C-ish sprintf */
12537 && q == fmtstart + 1 /* plain %d, not %....d */
12538 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12540 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12542 /* The argument has already gone through cBOOL, so the cast
12544 is_utf8 = (bool)va_arg(*args, int);
12545 elen = va_arg(*args, UV);
12546 /* if utf8 length is larger than 0x7ffff..., then it might
12547 * have been a signed value that wrapped */
12548 if (elen > ((~(STRLEN)0) >> 1)) {
12549 assert(0); /* in DEBUGGING build we want to crash */
12550 elen = 0; /* otherwise we want to treat this as an empty string */
12552 eptr = va_arg(*args, char *);
12553 q += sizeof(UTF8f) - 2;
12560 goto get_int_arg_val;
12571 goto get_int_arg_val;
12576 goto get_int_arg_val;
12587 goto get_int_arg_val;
12602 esignbuf[esignlen++] = plus;
12605 /* initialise the vector string to iterate over */
12607 vecsv = args ? va_arg(*args, SV*) : argsv;
12609 /* if this is a version object, we need to convert
12610 * back into v-string notation and then let the
12611 * vectorize happen normally
12613 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12614 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12615 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12616 "vector argument not supported with alpha versions");
12620 vecstr = (U8*)SvPV_const(vecsv,veclen);
12621 vecsv = sv_newmortal();
12622 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12626 vecstr = (U8*)SvPV_const(vecsv, veclen);
12627 vec_utf8 = DO_UTF8(vecsv);
12629 /* This is the re-entry point for when we're iterating
12630 * over the individual characters of a vector arg */
12633 goto done_valid_conversion;
12635 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12645 /* test arg for inf/nan. This can trigger an unwanted
12646 * 'str' overload, so manually force 'num' overload first
12650 if (UNLIKELY(SvAMAGIC(argsv)))
12651 argsv = sv_2num(argsv);
12652 if (UNLIKELY(isinfnansv(argsv)))
12653 goto handle_infnan_argsv;
12657 /* signed int type */
12662 case 'c': iv = (char)va_arg(*args, int); break;
12663 case 'h': iv = (short)va_arg(*args, int); break;
12664 case 'l': iv = va_arg(*args, long); break;
12665 case 'V': iv = va_arg(*args, IV); break;
12666 case 'z': iv = va_arg(*args, SSize_t); break;
12667 #ifdef HAS_PTRDIFF_T
12668 case 't': iv = va_arg(*args, ptrdiff_t); break;
12670 default: iv = va_arg(*args, int); break;
12671 case 'j': iv = (IV) va_arg(*args, PERL_INTMAX_T); break;
12674 iv = va_arg(*args, Quad_t); break;
12681 /* assign to tiv then cast to iv to work around
12682 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12683 IV tiv = SvIV_nomg(argsv);
12685 case 'c': iv = (char)tiv; break;
12686 case 'h': iv = (short)tiv; break;
12687 case 'l': iv = (long)tiv; break;
12689 default: iv = tiv; break;
12692 iv = (Quad_t)tiv; break;
12699 /* now convert iv to uv */
12703 esignbuf[esignlen++] = plus;
12706 /* Using 0- here to silence bogus warning from MS VC */
12707 uv = (UV) (0 - (UV) iv);
12708 esignbuf[esignlen++] = '-';
12712 /* unsigned int type */
12715 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12717 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12719 case 'l': uv = va_arg(*args, unsigned long); break;
12720 case 'V': uv = va_arg(*args, UV); break;
12721 case 'z': uv = va_arg(*args, Size_t); break;
12722 #ifdef HAS_PTRDIFF_T
12723 /* will sign extend, but there is no
12724 * uptrdiff_t, so oh well */
12725 case 't': uv = va_arg(*args, ptrdiff_t); break;
12727 case 'j': uv = (UV) va_arg(*args, PERL_UINTMAX_T); break;
12728 default: uv = va_arg(*args, unsigned); break;
12731 uv = va_arg(*args, Uquad_t); break;
12738 /* assign to tiv then cast to iv to work around
12739 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12740 UV tuv = SvUV_nomg(argsv);
12742 case 'c': uv = (unsigned char)tuv; break;
12743 case 'h': uv = (unsigned short)tuv; break;
12744 case 'l': uv = (unsigned long)tuv; break;
12746 default: uv = tuv; break;
12749 uv = (Uquad_t)tuv; break;
12760 char *ptr = ebuf + sizeof ebuf;
12767 const char * const p =
12768 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12773 } while (uv >>= 4);
12774 if (alt && *ptr != '0') {
12775 esignbuf[esignlen++] = '0';
12776 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12783 *--ptr = '0' + dig;
12784 } while (uv >>= 3);
12785 if (alt && *ptr != '0')
12791 *--ptr = '0' + dig;
12792 } while (uv >>= 1);
12793 if (alt && *ptr != '0') {
12794 esignbuf[esignlen++] = '0';
12795 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12800 /* special-case: base 1 indicates a 'c' format:
12801 * we use the common code for extracting a uv,
12802 * but handle that value differently here than
12803 * all the other int types */
12805 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12808 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12810 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12815 ebuf[0] = (char)uv;
12820 default: /* it had better be ten or less */
12823 *--ptr = '0' + dig;
12824 } while (uv /= base);
12827 elen = (ebuf + sizeof ebuf) - ptr;
12831 zeros = precis - elen;
12832 else if (precis == 0 && elen == 1 && *eptr == '0'
12833 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12836 /* a precision nullifies the 0 flag. */
12842 /* FLOATING POINT */
12845 c = 'f'; /* maybe %F isn't supported here */
12847 case 'e': case 'E':
12849 case 'g': case 'G':
12850 case 'a': case 'A':
12853 STRLEN float_need; /* what PL_efloatsize needs to become */
12854 bool hexfp; /* hexadecimal floating point? */
12856 vcatpvfn_long_double_t fv;
12859 /* This is evil, but floating point is even more evil */
12861 /* for SV-style calling, we can only get NV
12862 for C-style calling, we assume %f is double;
12863 for simplicity we allow any of %Lf, %llf, %qf for long double
12867 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12871 /* [perl #20339] - we should accept and ignore %lf rather than die */
12875 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12876 intsize = args ? 0 : 'q';
12880 #if defined(HAS_LONG_DOUBLE)
12893 /* Now we need (long double) if intsize == 'q', else (double). */
12895 /* Note: do not pull NVs off the va_list with va_arg()
12896 * (pull doubles instead) because if you have a build
12897 * with long doubles, you would always be pulling long
12898 * doubles, which would badly break anyone using only
12899 * doubles (i.e. the majority of builds). In other
12900 * words, you cannot mix doubles and long doubles.
12901 * The only case where you can pull off long doubles
12902 * is when the format specifier explicitly asks so with
12904 #ifdef USE_QUADMATH
12905 fv = intsize == 'q' ?
12906 va_arg(*args, NV) : va_arg(*args, double);
12908 #elif LONG_DOUBLESIZE > DOUBLESIZE
12909 if (intsize == 'q') {
12910 fv = va_arg(*args, long double);
12913 nv = va_arg(*args, double);
12914 VCATPVFN_NV_TO_FV(nv, fv);
12917 nv = va_arg(*args, double);
12924 /* we jump here if an int-ish format encountered an
12925 * infinite/Nan argsv. After setting nv/fv, it falls
12926 * into the isinfnan block which follows */
12927 handle_infnan_argsv:
12928 nv = SvNV_nomg(argsv);
12929 VCATPVFN_NV_TO_FV(nv, fv);
12932 if (Perl_isinfnan(nv)) {
12934 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12935 SvNV_nomg(argsv), (int)c);
12937 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12946 /* special-case "%.0f" */
12950 && !(width || left || plus || alt)
12953 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12957 /* Determine the buffer size needed for the various
12958 * floating-point formats.
12960 * The basic possibilities are:
12963 * %f 1111111.123456789
12964 * %e 1.111111123e+06
12965 * %a 0x1.0f4471f9bp+20
12967 * %g 1.11111112e+15
12969 * where P is the value of the precision in the format, or 6
12970 * if not specified. Note the two possible output formats of
12971 * %g; in both cases the number of significant digits is <=
12974 * For most of the format types the maximum buffer size needed
12975 * is precision, plus: any leading 1 or 0x1, the radix
12976 * point, and an exponent. The difficult one is %f: for a
12977 * large positive exponent it can have many leading digits,
12978 * which needs to be calculated specially. Also %a is slightly
12979 * different in that in the absence of a specified precision,
12980 * it uses as many digits as necessary to distinguish
12981 * different values.
12983 * First, here are the constant bits. For ease of calculation
12984 * we over-estimate the needed buffer size, for example by
12985 * assuming all formats have an exponent and a leading 0x1.
12987 * Also for production use, add a little extra overhead for
12988 * safety's sake. Under debugging don't, as it means we're
12989 * more likely to quickly spot issues during development.
12992 float_need = 1 /* possible unary minus */
12993 + 4 /* "0x1" plus very unlikely carry */
12994 + 1 /* default radix point '.' */
12995 + 2 /* "e-", "p+" etc */
12996 + 6 /* exponent: up to 16383 (quad fp) */
12998 + 20 /* safety net */
13003 /* determine the radix point len, e.g. length(".") in "1.2" */
13004 #ifdef USE_LOCALE_NUMERIC
13005 /* note that we may either explicitly use PL_numeric_radix_sv
13006 * below, or implicitly, via an snprintf() variant.
13007 * Note also things like ps_AF.utf8 which has
13008 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
13009 if (! have_in_lc_numeric) {
13010 in_lc_numeric = IN_LC(LC_NUMERIC);
13011 have_in_lc_numeric = TRUE;
13014 if (in_lc_numeric) {
13015 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, {
13016 /* this can't wrap unless PL_numeric_radix_sv is a string
13017 * consuming virtually all the 32-bit or 64-bit address
13020 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
13022 /* floating-point formats only get utf8 if the radix point
13023 * is utf8. All other characters in the string are < 128
13024 * and so can be safely appended to both a non-utf8 and utf8
13026 * Note that this will convert the output to utf8 even if
13027 * the radix point didn't get output.
13029 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
13030 sv_utf8_upgrade(sv);
13039 if (isALPHA_FOLD_EQ(c, 'f')) {
13040 /* Determine how many digits before the radix point
13041 * might be emitted. frexp() (or frexpl) has some
13042 * unspecified behaviour for nan/inf/-inf, so lucky we've
13043 * already handled them above */
13045 int i = PERL_INT_MIN;
13046 (void)Perl_frexp((NV)fv, &i);
13047 if (i == PERL_INT_MIN)
13048 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13051 digits = BIT_DIGITS(i);
13052 /* this can't overflow. 'digits' will only be a few
13053 * thousand even for the largest floating-point types.
13054 * And up until now float_need is just some small
13055 * constants plus radix len, which can't be in
13056 * overflow territory unless the radix SV is consuming
13057 * over 1/2 the address space */
13058 assert(float_need < ((STRLEN)~0) - digits);
13059 float_need += digits;
13062 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13065 /* %a in the absence of precision may print as many
13066 * digits as needed to represent the entire mantissa
13068 * This estimate seriously overshoots in most cases,
13069 * but better the undershooting. Firstly, all bytes
13070 * of the NV are not mantissa, some of them are
13071 * exponent. Secondly, for the reasonably common
13072 * long doubles case, the "80-bit extended", two
13073 * or six bytes of the NV are unused. Also, we'll
13074 * still pick up an extra +6 from the default
13075 * precision calculation below. */
13077 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13078 /* For the "double double", we need more.
13079 * Since each double has their own exponent, the
13080 * doubles may float (haha) rather far from each
13081 * other, and the number of required bits is much
13082 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13083 * See the definition of DOUBLEDOUBLE_MAXBITS.
13085 * Need 2 hexdigits for each byte. */
13086 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13088 NVSIZE * 2; /* 2 hexdigits for each byte */
13090 /* see "this can't overflow" comment above */
13091 assert(float_need < ((STRLEN)~0) - digits);
13092 float_need += digits;
13095 /* special-case "%.<number>g" if it will fit in ebuf */
13097 && precis /* See earlier comment about buggy Gconvert
13098 when digits, aka precis, is 0 */
13100 /* check, in manner not involving wrapping, that it will
13102 && float_need < sizeof(ebuf)
13103 && sizeof(ebuf) - float_need > precis
13104 && !(width || left || plus || alt)
13108 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13109 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis)
13111 elen = strlen(ebuf);
13118 STRLEN pr = has_precis ? precis : 6; /* known default */
13119 /* this probably can't wrap, since precis is limited
13120 * to 1/4 address space size, but better safe than sorry
13122 if (float_need >= ((STRLEN)~0) - pr)
13123 croak_memory_wrap();
13127 if (float_need < width)
13128 float_need = width;
13130 if (float_need > INT_MAX) {
13131 /* snprintf() returns an int, and we use that return value,
13132 so die horribly if the expected size is too large for int
13134 Perl_croak(aTHX_ "Numeric format result too large");
13137 if (PL_efloatsize <= float_need) {
13138 /* PL_efloatbuf should be at least 1 greater than
13139 * float_need to allow a trailing \0 to be returned by
13140 * snprintf(). If we need to grow, overgrow for the
13141 * benefit of future generations */
13142 const STRLEN extra = 0x20;
13143 if (float_need >= ((STRLEN)~0) - extra)
13144 croak_memory_wrap();
13145 float_need += extra;
13146 Safefree(PL_efloatbuf);
13147 PL_efloatsize = float_need;
13148 Newx(PL_efloatbuf, PL_efloatsize, char);
13149 PL_efloatbuf[0] = '\0';
13152 if (UNLIKELY(hexfp)) {
13153 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13154 nv, fv, has_precis, precis, width,
13155 alt, plus, left, fill, in_lc_numeric);
13158 char *ptr = ebuf + sizeof ebuf;
13161 #if defined(USE_QUADMATH)
13162 if (intsize == 'q') {
13166 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13167 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13168 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13169 * not USE_LONG_DOUBLE and NVff. In other words,
13170 * this needs to work without USE_LONG_DOUBLE. */
13171 if (intsize == 'q') {
13172 /* Copy the one or more characters in a long double
13173 * format before the 'base' ([efgEFG]) character to
13174 * the format string. */
13175 static char const ldblf[] = PERL_PRIfldbl;
13176 char const *p = ldblf + sizeof(ldblf) - 3;
13177 while (p >= ldblf) { *--ptr = *p--; }
13182 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13187 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13199 /* No taint. Otherwise we are in the strange situation
13200 * where printf() taints but print($float) doesn't.
13203 /* hopefully the above makes ptr a very constrained format
13204 * that is safe to use, even though it's not literal */
13205 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13206 #ifdef USE_QUADMATH
13208 const char* qfmt = quadmath_format_single(ptr);
13210 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13211 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13212 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13215 if ((IV)elen == -1) {
13218 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13223 #elif defined(HAS_LONG_DOUBLE)
13224 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13225 elen = ((intsize == 'q')
13226 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13227 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv))
13230 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13231 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13234 GCC_DIAG_RESTORE_STMT;
13237 eptr = PL_efloatbuf;
13241 /* Since floating-point formats do their own formatting and
13242 * padding, we skip the main block of code at the end of this
13243 * loop which handles appending eptr to sv, and do our own
13244 * stripped-down version */
13249 assert(elen >= width);
13251 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13253 goto done_valid_conversion;
13261 /* XXX ideally we should warn if any flags etc have been
13262 * set, e.g. "%-4.5n" */
13263 /* XXX if sv was originally non-utf8 with a char in the
13264 * range 0x80-0xff, then if it got upgraded, we should
13265 * calculate char len rather than byte len here */
13266 len = SvCUR(sv) - origlen;
13268 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13271 case 'c': *(va_arg(*args, char*)) = i; break;
13272 case 'h': *(va_arg(*args, short*)) = i; break;
13273 default: *(va_arg(*args, int*)) = i; break;
13274 case 'l': *(va_arg(*args, long*)) = i; break;
13275 case 'V': *(va_arg(*args, IV*)) = i; break;
13276 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13277 #ifdef HAS_PTRDIFF_T
13278 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13280 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13283 *(va_arg(*args, Quad_t*)) = i; break;
13291 Perl_croak_nocontext(
13292 "Missing argument for %%n in %s",
13293 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13294 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13296 goto done_valid_conversion;
13304 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13305 && ckWARN(WARN_PRINTF))
13307 SV * const msg = sv_newmortal();
13308 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13309 (PL_op->op_type == OP_PRTF) ? "" : "s");
13310 if (fmtstart < patend) {
13311 const char * const fmtend = q < patend ? q : patend;
13313 sv_catpvs(msg, "\"%");
13314 for (f = fmtstart; f < fmtend; f++) {
13316 sv_catpvn_nomg(msg, f, 1);
13318 Perl_sv_catpvf(aTHX_ msg,
13319 "\\%03" UVof, (UV)*f & 0xFF);
13322 sv_catpvs(msg, "\"");
13324 sv_catpvs(msg, "end of string");
13326 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13329 /* mangled format: output the '%', then continue from the
13330 * character following that */
13331 sv_catpvn_nomg(sv, fmtstart-1, 1);
13334 /* Any "redundant arg" warning from now onwards will probably
13335 * just be misleading, so don't bother. */
13336 no_redundant_warning = TRUE;
13337 continue; /* not "break" */
13340 if (is_utf8 != has_utf8) {
13343 sv_utf8_upgrade(sv);
13346 const STRLEN old_elen = elen;
13347 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13348 sv_utf8_upgrade(nsv);
13349 eptr = SvPVX_const(nsv);
13352 if (width) { /* fudge width (can't fudge elen) */
13353 width += elen - old_elen;
13360 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13363 STRLEN need, have, gap;
13367 /* signed value that's wrapped? */
13368 assert(elen <= ((~(STRLEN)0) >> 1));
13370 /* if zeros is non-zero, then it represents filler between
13371 * elen and precis. So adding elen and zeros together will
13372 * always be <= precis, and the addition can never wrap */
13373 assert(!zeros || (precis > elen && precis - elen == zeros));
13374 have = elen + zeros;
13376 if (have >= (((STRLEN)~0) - esignlen))
13377 croak_memory_wrap();
13380 need = (have > width ? have : width);
13383 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13384 croak_memory_wrap();
13385 need += (SvCUR(sv) + 1);
13392 for (i = 0; i < esignlen; i++)
13393 *s++ = esignbuf[i];
13394 for (i = zeros; i; i--)
13396 Copy(eptr, s, elen, char);
13398 for (i = gap; i; i--)
13403 for (i = 0; i < esignlen; i++)
13404 *s++ = esignbuf[i];
13409 for (i = gap; i; i--)
13411 for (i = 0; i < esignlen; i++)
13412 *s++ = esignbuf[i];
13415 for (i = zeros; i; i--)
13417 Copy(eptr, s, elen, char);
13422 SvCUR_set(sv, s - SvPVX_const(sv));
13430 if (vectorize && veclen) {
13431 /* we append the vector separator separately since %v isn't
13432 * very common: don't slow down the general case by adding
13433 * dotstrlen to need etc */
13434 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13436 goto vector; /* do next iteration */
13439 done_valid_conversion:
13442 S_warn_vcatpvfn_missing_argument(aTHX);
13445 /* Now that we've consumed all our printf format arguments (svix)
13446 * do we have things left on the stack that we didn't use?
13448 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13449 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13450 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13456 /* =========================================================================
13458 =head1 Cloning an interpreter
13462 All the macros and functions in this section are for the private use of
13463 the main function, perl_clone().
13465 The foo_dup() functions make an exact copy of an existing foo thingy.
13466 During the course of a cloning, a hash table is used to map old addresses
13467 to new addresses. The table is created and manipulated with the
13468 ptr_table_* functions.
13470 * =========================================================================*/
13473 #if defined(USE_ITHREADS)
13475 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13476 #ifndef GpREFCNT_inc
13477 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13481 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13482 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13483 If this changes, please unmerge ss_dup.
13484 Likewise, sv_dup_inc_multiple() relies on this fact. */
13485 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13486 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13487 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13488 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13489 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13490 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13491 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13492 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13493 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13494 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13495 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13496 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13497 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13499 /* clone a parser */
13502 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13506 PERL_ARGS_ASSERT_PARSER_DUP;
13511 /* look for it in the table first */
13512 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13516 /* create anew and remember what it is */
13517 Newxz(parser, 1, yy_parser);
13518 ptr_table_store(PL_ptr_table, proto, parser);
13520 /* XXX eventually, just Copy() most of the parser struct ? */
13522 parser->lex_brackets = proto->lex_brackets;
13523 parser->lex_casemods = proto->lex_casemods;
13524 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13525 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13526 parser->lex_casestack = savepvn(proto->lex_casestack,
13527 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13528 parser->lex_defer = proto->lex_defer;
13529 parser->lex_dojoin = proto->lex_dojoin;
13530 parser->lex_formbrack = proto->lex_formbrack;
13531 parser->lex_inpat = proto->lex_inpat;
13532 parser->lex_inwhat = proto->lex_inwhat;
13533 parser->lex_op = proto->lex_op;
13534 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13535 parser->lex_starts = proto->lex_starts;
13536 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13537 parser->multi_close = proto->multi_close;
13538 parser->multi_open = proto->multi_open;
13539 parser->multi_start = proto->multi_start;
13540 parser->multi_end = proto->multi_end;
13541 parser->preambled = proto->preambled;
13542 parser->lex_super_state = proto->lex_super_state;
13543 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13544 parser->lex_sub_op = proto->lex_sub_op;
13545 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13546 parser->linestr = sv_dup_inc(proto->linestr, param);
13547 parser->expect = proto->expect;
13548 parser->copline = proto->copline;
13549 parser->last_lop_op = proto->last_lop_op;
13550 parser->lex_state = proto->lex_state;
13551 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13552 /* rsfp_filters entries have fake IoDIRP() */
13553 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13554 parser->in_my = proto->in_my;
13555 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13556 parser->error_count = proto->error_count;
13557 parser->sig_elems = proto->sig_elems;
13558 parser->sig_optelems= proto->sig_optelems;
13559 parser->sig_slurpy = proto->sig_slurpy;
13560 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13563 char * const ols = SvPVX(proto->linestr);
13564 char * const ls = SvPVX(parser->linestr);
13566 parser->bufptr = ls + (proto->bufptr >= ols ?
13567 proto->bufptr - ols : 0);
13568 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13569 proto->oldbufptr - ols : 0);
13570 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13571 proto->oldoldbufptr - ols : 0);
13572 parser->linestart = ls + (proto->linestart >= ols ?
13573 proto->linestart - ols : 0);
13574 parser->last_uni = ls + (proto->last_uni >= ols ?
13575 proto->last_uni - ols : 0);
13576 parser->last_lop = ls + (proto->last_lop >= ols ?
13577 proto->last_lop - ols : 0);
13579 parser->bufend = ls + SvCUR(parser->linestr);
13582 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13585 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13586 Copy(proto->nexttype, parser->nexttype, 5, I32);
13587 parser->nexttoke = proto->nexttoke;
13589 /* XXX should clone saved_curcop here, but we aren't passed
13590 * proto_perl; so do it in perl_clone_using instead */
13596 /* duplicate a file handle */
13599 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13603 PERL_ARGS_ASSERT_FP_DUP;
13604 PERL_UNUSED_ARG(type);
13607 return (PerlIO*)NULL;
13609 /* look for it in the table first */
13610 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13614 /* create anew and remember what it is */
13615 #ifdef __amigaos4__
13616 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13618 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13620 ptr_table_store(PL_ptr_table, fp, ret);
13624 /* duplicate a directory handle */
13627 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13631 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13633 const Direntry_t *dirent;
13634 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13640 PERL_UNUSED_CONTEXT;
13641 PERL_ARGS_ASSERT_DIRP_DUP;
13646 /* look for it in the table first */
13647 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13651 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13653 PERL_UNUSED_ARG(param);
13657 /* open the current directory (so we can switch back) */
13658 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13660 /* chdir to our dir handle and open the present working directory */
13661 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13662 PerlDir_close(pwd);
13663 return (DIR *)NULL;
13665 /* Now we should have two dir handles pointing to the same dir. */
13667 /* Be nice to the calling code and chdir back to where we were. */
13668 /* XXX If this fails, then what? */
13669 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13671 /* We have no need of the pwd handle any more. */
13672 PerlDir_close(pwd);
13675 # define d_namlen(d) (d)->d_namlen
13677 # define d_namlen(d) strlen((d)->d_name)
13679 /* Iterate once through dp, to get the file name at the current posi-
13680 tion. Then step back. */
13681 pos = PerlDir_tell(dp);
13682 if ((dirent = PerlDir_read(dp))) {
13683 len = d_namlen(dirent);
13684 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13685 /* If the len is somehow magically longer than the
13686 * maximum length of the directory entry, even though
13687 * we could fit it in a buffer, we could not copy it
13688 * from the dirent. Bail out. */
13689 PerlDir_close(ret);
13692 if (len <= sizeof smallbuf) name = smallbuf;
13693 else Newx(name, len, char);
13694 Move(dirent->d_name, name, len, char);
13696 PerlDir_seek(dp, pos);
13698 /* Iterate through the new dir handle, till we find a file with the
13700 if (!dirent) /* just before the end */
13702 pos = PerlDir_tell(ret);
13703 if (PerlDir_read(ret)) continue; /* not there yet */
13704 PerlDir_seek(ret, pos); /* step back */
13708 const long pos0 = PerlDir_tell(ret);
13710 pos = PerlDir_tell(ret);
13711 if ((dirent = PerlDir_read(ret))) {
13712 if (len == (STRLEN)d_namlen(dirent)
13713 && memEQ(name, dirent->d_name, len)) {
13715 PerlDir_seek(ret, pos); /* step back */
13718 /* else we are not there yet; keep iterating */
13720 else { /* This is not meant to happen. The best we can do is
13721 reset the iterator to the beginning. */
13722 PerlDir_seek(ret, pos0);
13729 if (name && name != smallbuf)
13734 ret = win32_dirp_dup(dp, param);
13737 /* pop it in the pointer table */
13739 ptr_table_store(PL_ptr_table, dp, ret);
13744 /* duplicate a typeglob */
13747 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13751 PERL_ARGS_ASSERT_GP_DUP;
13755 /* look for it in the table first */
13756 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13760 /* create anew and remember what it is */
13762 ptr_table_store(PL_ptr_table, gp, ret);
13765 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13766 on Newxz() to do this for us. */
13767 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13768 ret->gp_io = io_dup_inc(gp->gp_io, param);
13769 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13770 ret->gp_av = av_dup_inc(gp->gp_av, param);
13771 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13772 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13773 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13774 ret->gp_cvgen = gp->gp_cvgen;
13775 ret->gp_line = gp->gp_line;
13776 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13780 /* duplicate a chain of magic */
13783 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13785 MAGIC *mgret = NULL;
13786 MAGIC **mgprev_p = &mgret;
13788 PERL_ARGS_ASSERT_MG_DUP;
13790 for (; mg; mg = mg->mg_moremagic) {
13793 if ((param->flags & CLONEf_JOIN_IN)
13794 && mg->mg_type == PERL_MAGIC_backref)
13795 /* when joining, we let the individual SVs add themselves to
13796 * backref as needed. */
13799 Newx(nmg, 1, MAGIC);
13801 mgprev_p = &(nmg->mg_moremagic);
13803 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13804 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13805 from the original commit adding Perl_mg_dup() - revision 4538.
13806 Similarly there is the annotation "XXX random ptr?" next to the
13807 assignment to nmg->mg_ptr. */
13810 /* FIXME for plugins
13811 if (nmg->mg_type == PERL_MAGIC_qr) {
13812 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13816 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13817 ? nmg->mg_type == PERL_MAGIC_backref
13818 /* The backref AV has its reference
13819 * count deliberately bumped by 1 */
13820 ? SvREFCNT_inc(av_dup_inc((const AV *)
13821 nmg->mg_obj, param))
13822 : sv_dup_inc(nmg->mg_obj, param)
13823 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13824 nmg->mg_type == PERL_MAGIC_regdata)
13826 : sv_dup(nmg->mg_obj, param);
13828 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13829 if (nmg->mg_len > 0) {
13830 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13831 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13832 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13834 AMT * const namtp = (AMT*)nmg->mg_ptr;
13835 sv_dup_inc_multiple((SV**)(namtp->table),
13836 (SV**)(namtp->table), NofAMmeth, param);
13839 else if (nmg->mg_len == HEf_SVKEY)
13840 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13842 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13843 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13849 #endif /* USE_ITHREADS */
13851 struct ptr_tbl_arena {
13852 struct ptr_tbl_arena *next;
13853 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13856 /* create a new pointer-mapping table */
13859 Perl_ptr_table_new(pTHX)
13862 PERL_UNUSED_CONTEXT;
13864 Newx(tbl, 1, PTR_TBL_t);
13865 tbl->tbl_max = 511;
13866 tbl->tbl_items = 0;
13867 tbl->tbl_arena = NULL;
13868 tbl->tbl_arena_next = NULL;
13869 tbl->tbl_arena_end = NULL;
13870 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13874 #define PTR_TABLE_HASH(ptr) \
13875 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13877 /* map an existing pointer using a table */
13879 STATIC PTR_TBL_ENT_t *
13880 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13882 PTR_TBL_ENT_t *tblent;
13883 const UV hash = PTR_TABLE_HASH(sv);
13885 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13887 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13888 for (; tblent; tblent = tblent->next) {
13889 if (tblent->oldval == sv)
13896 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13898 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13900 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13901 PERL_UNUSED_CONTEXT;
13903 return tblent ? tblent->newval : NULL;
13906 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13907 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13908 * the core's typical use of ptr_tables in thread cloning. */
13911 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13913 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13915 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13916 PERL_UNUSED_CONTEXT;
13919 tblent->newval = newsv;
13921 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13923 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13924 struct ptr_tbl_arena *new_arena;
13926 Newx(new_arena, 1, struct ptr_tbl_arena);
13927 new_arena->next = tbl->tbl_arena;
13928 tbl->tbl_arena = new_arena;
13929 tbl->tbl_arena_next = new_arena->array;
13930 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13933 tblent = tbl->tbl_arena_next++;
13935 tblent->oldval = oldsv;
13936 tblent->newval = newsv;
13937 tblent->next = tbl->tbl_ary[entry];
13938 tbl->tbl_ary[entry] = tblent;
13940 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13941 ptr_table_split(tbl);
13945 /* double the hash bucket size of an existing ptr table */
13948 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13950 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13951 const UV oldsize = tbl->tbl_max + 1;
13952 UV newsize = oldsize * 2;
13955 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13956 PERL_UNUSED_CONTEXT;
13958 Renew(ary, newsize, PTR_TBL_ENT_t*);
13959 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13960 tbl->tbl_max = --newsize;
13961 tbl->tbl_ary = ary;
13962 for (i=0; i < oldsize; i++, ary++) {
13963 PTR_TBL_ENT_t **entp = ary;
13964 PTR_TBL_ENT_t *ent = *ary;
13965 PTR_TBL_ENT_t **curentp;
13968 curentp = ary + oldsize;
13970 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13972 ent->next = *curentp;
13982 /* remove all the entries from a ptr table */
13983 /* Deprecated - will be removed post 5.14 */
13986 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13988 PERL_UNUSED_CONTEXT;
13989 if (tbl && tbl->tbl_items) {
13990 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13992 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13995 struct ptr_tbl_arena *next = arena->next;
14001 tbl->tbl_items = 0;
14002 tbl->tbl_arena = NULL;
14003 tbl->tbl_arena_next = NULL;
14004 tbl->tbl_arena_end = NULL;
14008 /* clear and free a ptr table */
14011 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
14013 struct ptr_tbl_arena *arena;
14015 PERL_UNUSED_CONTEXT;
14021 arena = tbl->tbl_arena;
14024 struct ptr_tbl_arena *next = arena->next;
14030 Safefree(tbl->tbl_ary);
14034 #if defined(USE_ITHREADS)
14037 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14039 PERL_ARGS_ASSERT_RVPV_DUP;
14041 assert(!isREGEXP(sstr));
14043 if (SvWEAKREF(sstr)) {
14044 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14045 if (param->flags & CLONEf_JOIN_IN) {
14046 /* if joining, we add any back references individually rather
14047 * than copying the whole backref array */
14048 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14052 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14054 else if (SvPVX_const(sstr)) {
14055 /* Has something there */
14057 /* Normal PV - clone whole allocated space */
14058 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14059 /* sstr may not be that normal, but actually copy on write.
14060 But we are a true, independent SV, so: */
14064 /* Special case - not normally malloced for some reason */
14065 if (isGV_with_GP(sstr)) {
14066 /* Don't need to do anything here. */
14068 else if ((SvIsCOW(sstr))) {
14069 /* A "shared" PV - clone it as "shared" PV */
14071 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14075 /* Some other special case - random pointer */
14076 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14081 /* Copy the NULL */
14082 SvPV_set(dstr, NULL);
14086 /* duplicate a list of SVs. source and dest may point to the same memory. */
14088 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14089 SSize_t items, CLONE_PARAMS *const param)
14091 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14093 while (items-- > 0) {
14094 *dest++ = sv_dup_inc(*source++, param);
14100 /* duplicate an SV of any type (including AV, HV etc) */
14103 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14108 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14110 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14111 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14116 /* look for it in the table first */
14117 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14121 if(param->flags & CLONEf_JOIN_IN) {
14122 /** We are joining here so we don't want do clone
14123 something that is bad **/
14124 if (SvTYPE(sstr) == SVt_PVHV) {
14125 const HEK * const hvname = HvNAME_HEK(sstr);
14127 /** don't clone stashes if they already exist **/
14128 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14129 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14130 ptr_table_store(PL_ptr_table, sstr, dstr);
14134 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14135 HV *stash = GvSTASH(sstr);
14136 const HEK * hvname;
14137 if (stash && (hvname = HvNAME_HEK(stash))) {
14138 /** don't clone GVs if they already exist **/
14140 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14141 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14143 stash, GvNAME(sstr),
14149 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14150 ptr_table_store(PL_ptr_table, sstr, *svp);
14157 /* create anew and remember what it is */
14160 #ifdef DEBUG_LEAKING_SCALARS
14161 dstr->sv_debug_optype = sstr->sv_debug_optype;
14162 dstr->sv_debug_line = sstr->sv_debug_line;
14163 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14164 dstr->sv_debug_parent = (SV*)sstr;
14165 FREE_SV_DEBUG_FILE(dstr);
14166 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14169 ptr_table_store(PL_ptr_table, sstr, dstr);
14172 SvFLAGS(dstr) = SvFLAGS(sstr);
14173 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14174 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14177 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14178 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14179 (void*)PL_watch_pvx, SvPVX_const(sstr));
14182 /* don't clone objects whose class has asked us not to */
14184 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14190 switch (SvTYPE(sstr)) {
14192 SvANY(dstr) = NULL;
14195 SET_SVANY_FOR_BODYLESS_IV(dstr);
14197 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14199 SvIV_set(dstr, SvIVX(sstr));
14203 #if NVSIZE <= IVSIZE
14204 SET_SVANY_FOR_BODYLESS_NV(dstr);
14206 SvANY(dstr) = new_XNV();
14208 SvNV_set(dstr, SvNVX(sstr));
14212 /* These are all the types that need complex bodies allocating. */
14214 const svtype sv_type = SvTYPE(sstr);
14215 const struct body_details *const sv_type_details
14216 = bodies_by_type + sv_type;
14220 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14221 NOT_REACHED; /* NOTREACHED */
14237 assert(sv_type_details->body_size);
14238 if (sv_type_details->arena) {
14239 new_body_inline(new_body, sv_type);
14241 = (void*)((char*)new_body - sv_type_details->offset);
14243 new_body = new_NOARENA(sv_type_details);
14247 SvANY(dstr) = new_body;
14250 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14251 ((char*)SvANY(dstr)) + sv_type_details->offset,
14252 sv_type_details->copy, char);
14254 Copy(((char*)SvANY(sstr)),
14255 ((char*)SvANY(dstr)),
14256 sv_type_details->body_size + sv_type_details->offset, char);
14259 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14260 && !isGV_with_GP(dstr)
14262 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14263 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14265 /* The Copy above means that all the source (unduplicated) pointers
14266 are now in the destination. We can check the flags and the
14267 pointers in either, but it's possible that there's less cache
14268 missing by always going for the destination.
14269 FIXME - instrument and check that assumption */
14270 if (sv_type >= SVt_PVMG) {
14272 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14273 if (SvOBJECT(dstr) && SvSTASH(dstr))
14274 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14275 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14278 /* The cast silences a GCC warning about unhandled types. */
14279 switch ((int)sv_type) {
14290 /* FIXME for plugins */
14291 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14294 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14295 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14296 LvTARG(dstr) = dstr;
14297 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14298 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14300 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14301 if (isREGEXP(sstr)) goto duprex;
14304 /* non-GP case already handled above */
14305 if(isGV_with_GP(sstr)) {
14306 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14307 /* Don't call sv_add_backref here as it's going to be
14308 created as part of the magic cloning of the symbol
14309 table--unless this is during a join and the stash
14310 is not actually being cloned. */
14311 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14312 at the point of this comment. */
14313 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14314 if (param->flags & CLONEf_JOIN_IN)
14315 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14316 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14317 (void)GpREFCNT_inc(GvGP(dstr));
14321 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14322 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14323 /* I have no idea why fake dirp (rsfps)
14324 should be treated differently but otherwise
14325 we end up with leaks -- sky*/
14326 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14327 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14328 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14330 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14331 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14332 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14333 if (IoDIRP(dstr)) {
14334 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14337 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14339 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14341 if (IoOFP(dstr) == IoIFP(sstr))
14342 IoOFP(dstr) = IoIFP(dstr);
14344 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14345 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14346 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14347 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14350 /* avoid cloning an empty array */
14351 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14352 SV **dst_ary, **src_ary;
14353 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14355 src_ary = AvARRAY((const AV *)sstr);
14356 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14357 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14358 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14359 AvALLOC((const AV *)dstr) = dst_ary;
14360 if (AvREAL((const AV *)sstr)) {
14361 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14365 while (items-- > 0)
14366 *dst_ary++ = sv_dup(*src_ary++, param);
14368 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14369 while (items-- > 0) {
14374 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14375 AvALLOC((const AV *)dstr) = (SV**)NULL;
14376 AvMAX( (const AV *)dstr) = -1;
14377 AvFILLp((const AV *)dstr) = -1;
14381 if (HvARRAY((const HV *)sstr)) {
14383 const bool sharekeys = !!HvSHAREKEYS(sstr);
14384 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14385 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14387 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14388 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14390 HvARRAY(dstr) = (HE**)darray;
14391 while (i <= sxhv->xhv_max) {
14392 const HE * const source = HvARRAY(sstr)[i];
14393 HvARRAY(dstr)[i] = source
14394 ? he_dup(source, sharekeys, param) : 0;
14398 const struct xpvhv_aux * const saux = HvAUX(sstr);
14399 struct xpvhv_aux * const daux = HvAUX(dstr);
14400 /* This flag isn't copied. */
14403 if (saux->xhv_name_count) {
14404 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14406 = saux->xhv_name_count < 0
14407 ? -saux->xhv_name_count
14408 : saux->xhv_name_count;
14409 HEK **shekp = sname + count;
14411 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14412 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14413 while (shekp-- > sname) {
14415 *dhekp = hek_dup(*shekp, param);
14419 daux->xhv_name_u.xhvnameu_name
14420 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14423 daux->xhv_name_count = saux->xhv_name_count;
14425 daux->xhv_aux_flags = saux->xhv_aux_flags;
14426 #ifdef PERL_HASH_RANDOMIZE_KEYS
14427 daux->xhv_rand = saux->xhv_rand;
14428 daux->xhv_last_rand = saux->xhv_last_rand;
14430 daux->xhv_riter = saux->xhv_riter;
14431 daux->xhv_eiter = saux->xhv_eiter
14432 ? he_dup(saux->xhv_eiter,
14433 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14434 /* backref array needs refcnt=2; see sv_add_backref */
14435 daux->xhv_backreferences =
14436 (param->flags & CLONEf_JOIN_IN)
14437 /* when joining, we let the individual GVs and
14438 * CVs add themselves to backref as
14439 * needed. This avoids pulling in stuff
14440 * that isn't required, and simplifies the
14441 * case where stashes aren't cloned back
14442 * if they already exist in the parent
14445 : saux->xhv_backreferences
14446 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14447 ? MUTABLE_AV(SvREFCNT_inc(
14448 sv_dup_inc((const SV *)
14449 saux->xhv_backreferences, param)))
14450 : MUTABLE_AV(sv_dup((const SV *)
14451 saux->xhv_backreferences, param))
14454 daux->xhv_mro_meta = saux->xhv_mro_meta
14455 ? mro_meta_dup(saux->xhv_mro_meta, param)
14458 /* Record stashes for possible cloning in Perl_clone(). */
14460 av_push(param->stashes, dstr);
14464 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14467 if (!(param->flags & CLONEf_COPY_STACKS)) {
14472 /* NOTE: not refcounted */
14473 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14474 hv_dup(CvSTASH(dstr), param);
14475 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14476 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14477 if (!CvISXSUB(dstr)) {
14479 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14481 CvSLABBED_off(dstr);
14482 } else if (CvCONST(dstr)) {
14483 CvXSUBANY(dstr).any_ptr =
14484 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14486 assert(!CvSLABBED(dstr));
14487 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14489 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14490 hek_dup(CvNAME_HEK((CV *)sstr), param);
14491 /* don't dup if copying back - CvGV isn't refcounted, so the
14492 * duped GV may never be freed. A bit of a hack! DAPM */
14494 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14496 ? gv_dup_inc(CvGV(sstr), param)
14497 : (param->flags & CLONEf_JOIN_IN)
14499 : gv_dup(CvGV(sstr), param);
14501 if (!CvISXSUB(sstr)) {
14502 PADLIST * padlist = CvPADLIST(sstr);
14504 padlist = padlist_dup(padlist, param);
14505 CvPADLIST_set(dstr, padlist);
14507 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14508 PoisonPADLIST(dstr);
14511 CvWEAKOUTSIDE(sstr)
14512 ? cv_dup( CvOUTSIDE(dstr), param)
14513 : cv_dup_inc(CvOUTSIDE(dstr), param);
14523 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14525 PERL_ARGS_ASSERT_SV_DUP_INC;
14526 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14530 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14532 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14533 PERL_ARGS_ASSERT_SV_DUP;
14535 /* Track every SV that (at least initially) had a reference count of 0.
14536 We need to do this by holding an actual reference to it in this array.
14537 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14538 (akin to the stashes hash, and the perl stack), we come unstuck if
14539 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14540 thread) is manipulated in a CLONE method, because CLONE runs before the
14541 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14542 (and fix things up by giving each a reference via the temps stack).
14543 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14544 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14545 before the walk of unreferenced happens and a reference to that is SV
14546 added to the temps stack. At which point we have the same SV considered
14547 to be in use, and free to be re-used. Not good.
14549 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14550 assert(param->unreferenced);
14551 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14557 /* duplicate a context */
14560 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14562 PERL_CONTEXT *ncxs;
14564 PERL_ARGS_ASSERT_CX_DUP;
14567 return (PERL_CONTEXT*)NULL;
14569 /* look for it in the table first */
14570 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14574 /* create anew and remember what it is */
14575 Newx(ncxs, max + 1, PERL_CONTEXT);
14576 ptr_table_store(PL_ptr_table, cxs, ncxs);
14577 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14580 PERL_CONTEXT * const ncx = &ncxs[ix];
14581 if (CxTYPE(ncx) == CXt_SUBST) {
14582 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14585 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14586 switch (CxTYPE(ncx)) {
14588 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14589 if(CxHASARGS(ncx)){
14590 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14592 ncx->blk_sub.savearray = NULL;
14594 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14595 ncx->blk_sub.prevcomppad);
14598 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14600 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14601 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14602 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14603 /* XXX what do do with cur_top_env ???? */
14605 case CXt_LOOP_LAZYSV:
14606 ncx->blk_loop.state_u.lazysv.end
14607 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14608 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14609 duplication code instead.
14610 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14611 actually being the same function, and (2) order
14612 equivalence of the two unions.
14613 We can assert the later [but only at run time :-(] */
14614 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14615 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14618 ncx->blk_loop.state_u.ary.ary
14619 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14621 case CXt_LOOP_LIST:
14622 case CXt_LOOP_LAZYIV:
14623 /* code common to all 'for' CXt_LOOP_* types */
14624 ncx->blk_loop.itersave =
14625 sv_dup_inc(ncx->blk_loop.itersave, param);
14626 if (CxPADLOOP(ncx)) {
14627 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14628 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14629 ncx->blk_loop.oldcomppad =
14630 (PAD*)ptr_table_fetch(PL_ptr_table,
14631 ncx->blk_loop.oldcomppad);
14632 ncx->blk_loop.itervar_u.svp =
14633 &CX_CURPAD_SV(ncx->blk_loop, off);
14636 /* this copies the GV if CXp_FOR_GV, or the SV for an
14637 * alias (for \$x (...)) - relies on gv_dup being the
14638 * same as sv_dup */
14639 ncx->blk_loop.itervar_u.gv
14640 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14644 case CXt_LOOP_PLAIN:
14647 ncx->blk_format.prevcomppad =
14648 (PAD*)ptr_table_fetch(PL_ptr_table,
14649 ncx->blk_format.prevcomppad);
14650 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14651 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14652 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14656 ncx->blk_givwhen.defsv_save =
14657 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14670 /* duplicate a stack info structure */
14673 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14677 PERL_ARGS_ASSERT_SI_DUP;
14680 return (PERL_SI*)NULL;
14682 /* look for it in the table first */
14683 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14687 /* create anew and remember what it is */
14688 Newx(nsi, 1, PERL_SI);
14689 ptr_table_store(PL_ptr_table, si, nsi);
14691 nsi->si_stack = av_dup_inc(si->si_stack, param);
14692 nsi->si_cxix = si->si_cxix;
14693 nsi->si_cxsubix = si->si_cxsubix;
14694 nsi->si_cxmax = si->si_cxmax;
14695 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14696 nsi->si_type = si->si_type;
14697 nsi->si_prev = si_dup(si->si_prev, param);
14698 nsi->si_next = si_dup(si->si_next, param);
14699 nsi->si_markoff = si->si_markoff;
14700 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14701 nsi->si_stack_hwm = 0;
14707 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14708 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14709 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14710 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14711 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14712 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14713 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14714 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14715 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14716 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14717 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14718 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14719 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14720 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14721 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14722 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14725 #define pv_dup_inc(p) SAVEPV(p)
14726 #define pv_dup(p) SAVEPV(p)
14727 #define svp_dup_inc(p,pp) any_dup(p,pp)
14729 /* map any object to the new equivent - either something in the
14730 * ptr table, or something in the interpreter structure
14734 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14738 PERL_ARGS_ASSERT_ANY_DUP;
14741 return (void*)NULL;
14743 /* look for it in the table first */
14744 ret = ptr_table_fetch(PL_ptr_table, v);
14748 /* see if it is part of the interpreter structure */
14749 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14750 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14758 /* duplicate the save stack */
14761 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14764 ANY * const ss = proto_perl->Isavestack;
14765 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14766 I32 ix = proto_perl->Isavestack_ix;
14779 void (*dptr) (void*);
14780 void (*dxptr) (pTHX_ void*);
14782 PERL_ARGS_ASSERT_SS_DUP;
14784 Newx(nss, max, ANY);
14787 const UV uv = POPUV(ss,ix);
14788 const U8 type = (U8)uv & SAVE_MASK;
14790 TOPUV(nss,ix) = uv;
14792 case SAVEt_CLEARSV:
14793 case SAVEt_CLEARPADRANGE:
14795 case SAVEt_HELEM: /* hash element */
14796 case SAVEt_SV: /* scalar reference */
14797 sv = (const SV *)POPPTR(ss,ix);
14798 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14800 case SAVEt_ITEM: /* normal string */
14801 case SAVEt_GVSV: /* scalar slot in GV */
14802 sv = (const SV *)POPPTR(ss,ix);
14803 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14804 if (type == SAVEt_SV)
14808 case SAVEt_MORTALIZESV:
14809 case SAVEt_READONLY_OFF:
14810 sv = (const SV *)POPPTR(ss,ix);
14811 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14813 case SAVEt_FREEPADNAME:
14814 ptr = POPPTR(ss,ix);
14815 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14816 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14818 case SAVEt_SHARED_PVREF: /* char* in shared space */
14819 c = (char*)POPPTR(ss,ix);
14820 TOPPTR(nss,ix) = savesharedpv(c);
14821 ptr = POPPTR(ss,ix);
14822 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14824 case SAVEt_GENERIC_SVREF: /* generic sv */
14825 case SAVEt_SVREF: /* scalar reference */
14826 sv = (const SV *)POPPTR(ss,ix);
14827 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14828 if (type == SAVEt_SVREF)
14829 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14830 ptr = POPPTR(ss,ix);
14831 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14833 case SAVEt_GVSLOT: /* any slot in GV */
14834 sv = (const SV *)POPPTR(ss,ix);
14835 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14836 ptr = POPPTR(ss,ix);
14837 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14838 sv = (const SV *)POPPTR(ss,ix);
14839 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14841 case SAVEt_HV: /* hash reference */
14842 case SAVEt_AV: /* array reference */
14843 sv = (const SV *) POPPTR(ss,ix);
14844 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14846 case SAVEt_COMPPAD:
14848 sv = (const SV *) POPPTR(ss,ix);
14849 TOPPTR(nss,ix) = sv_dup(sv, param);
14851 case SAVEt_INT: /* int reference */
14852 ptr = POPPTR(ss,ix);
14853 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14854 intval = (int)POPINT(ss,ix);
14855 TOPINT(nss,ix) = intval;
14857 case SAVEt_LONG: /* long reference */
14858 ptr = POPPTR(ss,ix);
14859 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14860 longval = (long)POPLONG(ss,ix);
14861 TOPLONG(nss,ix) = longval;
14863 case SAVEt_I32: /* I32 reference */
14864 ptr = POPPTR(ss,ix);
14865 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14867 TOPINT(nss,ix) = i;
14869 case SAVEt_IV: /* IV reference */
14870 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14871 ptr = POPPTR(ss,ix);
14872 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14874 TOPIV(nss,ix) = iv;
14876 case SAVEt_TMPSFLOOR:
14878 TOPIV(nss,ix) = iv;
14880 case SAVEt_HPTR: /* HV* reference */
14881 case SAVEt_APTR: /* AV* reference */
14882 case SAVEt_SPTR: /* SV* reference */
14883 ptr = POPPTR(ss,ix);
14884 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14885 sv = (const SV *)POPPTR(ss,ix);
14886 TOPPTR(nss,ix) = sv_dup(sv, param);
14888 case SAVEt_VPTR: /* random* reference */
14889 ptr = POPPTR(ss,ix);
14890 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14892 case SAVEt_INT_SMALL:
14893 case SAVEt_I32_SMALL:
14894 case SAVEt_I16: /* I16 reference */
14895 case SAVEt_I8: /* I8 reference */
14897 ptr = POPPTR(ss,ix);
14898 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14900 case SAVEt_GENERIC_PVREF: /* generic char* */
14901 case SAVEt_PPTR: /* char* reference */
14902 ptr = POPPTR(ss,ix);
14903 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14904 c = (char*)POPPTR(ss,ix);
14905 TOPPTR(nss,ix) = pv_dup(c);
14907 case SAVEt_GP: /* scalar reference */
14908 gp = (GP*)POPPTR(ss,ix);
14909 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14910 (void)GpREFCNT_inc(gp);
14911 gv = (const GV *)POPPTR(ss,ix);
14912 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14915 ptr = POPPTR(ss,ix);
14916 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14917 /* these are assumed to be refcounted properly */
14919 switch (((OP*)ptr)->op_type) {
14921 case OP_LEAVESUBLV:
14925 case OP_LEAVEWRITE:
14926 TOPPTR(nss,ix) = ptr;
14929 (void) OpREFCNT_inc(o);
14933 TOPPTR(nss,ix) = NULL;
14938 TOPPTR(nss,ix) = NULL;
14940 case SAVEt_FREECOPHH:
14941 ptr = POPPTR(ss,ix);
14942 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14944 case SAVEt_ADELETE:
14945 av = (const AV *)POPPTR(ss,ix);
14946 TOPPTR(nss,ix) = av_dup_inc(av, param);
14948 TOPINT(nss,ix) = i;
14951 hv = (const HV *)POPPTR(ss,ix);
14952 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14954 TOPINT(nss,ix) = i;
14957 c = (char*)POPPTR(ss,ix);
14958 TOPPTR(nss,ix) = pv_dup_inc(c);
14960 case SAVEt_STACK_POS: /* Position on Perl stack */
14962 TOPINT(nss,ix) = i;
14964 case SAVEt_DESTRUCTOR:
14965 ptr = POPPTR(ss,ix);
14966 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14967 dptr = POPDPTR(ss,ix);
14968 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14969 any_dup(FPTR2DPTR(void *, dptr),
14972 case SAVEt_DESTRUCTOR_X:
14973 ptr = POPPTR(ss,ix);
14974 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14975 dxptr = POPDXPTR(ss,ix);
14976 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14977 any_dup(FPTR2DPTR(void *, dxptr),
14980 case SAVEt_REGCONTEXT:
14982 ix -= uv >> SAVE_TIGHT_SHIFT;
14984 case SAVEt_AELEM: /* array element */
14985 sv = (const SV *)POPPTR(ss,ix);
14986 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14988 TOPIV(nss,ix) = iv;
14989 av = (const AV *)POPPTR(ss,ix);
14990 TOPPTR(nss,ix) = av_dup_inc(av, param);
14993 ptr = POPPTR(ss,ix);
14994 TOPPTR(nss,ix) = ptr;
14997 ptr = POPPTR(ss,ix);
14998 ptr = cophh_copy((COPHH*)ptr);
14999 TOPPTR(nss,ix) = ptr;
15001 TOPINT(nss,ix) = i;
15002 if (i & HINT_LOCALIZE_HH) {
15003 hv = (const HV *)POPPTR(ss,ix);
15004 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
15007 case SAVEt_PADSV_AND_MORTALIZE:
15008 longval = (long)POPLONG(ss,ix);
15009 TOPLONG(nss,ix) = longval;
15010 ptr = POPPTR(ss,ix);
15011 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
15012 sv = (const SV *)POPPTR(ss,ix);
15013 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
15015 case SAVEt_SET_SVFLAGS:
15017 TOPINT(nss,ix) = i;
15019 TOPINT(nss,ix) = i;
15020 sv = (const SV *)POPPTR(ss,ix);
15021 TOPPTR(nss,ix) = sv_dup(sv, param);
15023 case SAVEt_COMPILE_WARNINGS:
15024 ptr = POPPTR(ss,ix);
15025 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
15028 ptr = POPPTR(ss,ix);
15029 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
15033 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
15041 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15042 * flag to the result. This is done for each stash before cloning starts,
15043 * so we know which stashes want their objects cloned */
15046 do_mark_cloneable_stash(pTHX_ SV *const sv)
15048 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15050 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15051 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15052 if (cloner && GvCV(cloner)) {
15059 mXPUSHs(newSVhek(hvname));
15061 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15068 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15076 =for apidoc perl_clone
15078 Create and return a new interpreter by cloning the current one.
15080 C<perl_clone> takes these flags as parameters:
15082 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15083 without it we only clone the data and zero the stacks,
15084 with it we copy the stacks and the new perl interpreter is
15085 ready to run at the exact same point as the previous one.
15086 The pseudo-fork code uses C<COPY_STACKS> while the
15087 threads->create doesn't.
15089 C<CLONEf_KEEP_PTR_TABLE> -
15090 C<perl_clone> keeps a ptr_table with the pointer of the old
15091 variable as a key and the new variable as a value,
15092 this allows it to check if something has been cloned and not
15093 clone it again, but rather just use the value and increase the
15095 If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill the ptr_table
15096 using the function S<C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>>.
15097 A reason to keep it around is if you want to dup some of your own
15098 variables which are outside the graph that perl scans.
15100 C<CLONEf_CLONE_HOST> -
15101 This is a win32 thing, it is ignored on unix, it tells perl's
15102 win32host code (which is c++) to clone itself, this is needed on
15103 win32 if you want to run two threads at the same time,
15104 if you just want to do some stuff in a separate perl interpreter
15105 and then throw it away and return to the original one,
15106 you don't need to do anything.
15111 /* XXX the above needs expanding by someone who actually understands it ! */
15112 EXTERN_C PerlInterpreter *
15113 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15116 perl_clone(PerlInterpreter *proto_perl, UV flags)
15119 #ifdef PERL_IMPLICIT_SYS
15121 PERL_ARGS_ASSERT_PERL_CLONE;
15123 /* perlhost.h so we need to call into it
15124 to clone the host, CPerlHost should have a c interface, sky */
15126 #ifndef __amigaos4__
15127 if (flags & CLONEf_CLONE_HOST) {
15128 return perl_clone_host(proto_perl,flags);
15131 return perl_clone_using(proto_perl, flags,
15133 proto_perl->IMemShared,
15134 proto_perl->IMemParse,
15136 proto_perl->IStdIO,
15140 proto_perl->IProc);
15144 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15145 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15146 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15147 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15148 struct IPerlDir* ipD, struct IPerlSock* ipS,
15149 struct IPerlProc* ipP)
15151 /* XXX many of the string copies here can be optimized if they're
15152 * constants; they need to be allocated as common memory and just
15153 * their pointers copied. */
15156 CLONE_PARAMS clone_params;
15157 CLONE_PARAMS* const param = &clone_params;
15159 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15161 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15162 #else /* !PERL_IMPLICIT_SYS */
15164 CLONE_PARAMS clone_params;
15165 CLONE_PARAMS* param = &clone_params;
15166 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15168 PERL_ARGS_ASSERT_PERL_CLONE;
15169 #endif /* PERL_IMPLICIT_SYS */
15171 /* for each stash, determine whether its objects should be cloned */
15172 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15173 PERL_SET_THX(my_perl);
15176 PoisonNew(my_perl, 1, PerlInterpreter);
15179 PL_defstash = NULL; /* may be used by perl malloc() */
15182 PL_scopestack_name = 0;
15184 PL_savestack_ix = 0;
15185 PL_savestack_max = -1;
15186 PL_sig_pending = 0;
15188 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15189 Zero(&PL_padname_undef, 1, PADNAME);
15190 Zero(&PL_padname_const, 1, PADNAME);
15191 # ifdef DEBUG_LEAKING_SCALARS
15192 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15194 # ifdef PERL_TRACE_OPS
15195 Zero(PL_op_exec_cnt, OP_max+2, UV);
15197 #else /* !DEBUGGING */
15198 Zero(my_perl, 1, PerlInterpreter);
15199 #endif /* DEBUGGING */
15201 #ifdef PERL_IMPLICIT_SYS
15202 /* host pointers */
15204 PL_MemShared = ipMS;
15205 PL_MemParse = ipMP;
15212 #endif /* PERL_IMPLICIT_SYS */
15215 param->flags = flags;
15216 /* Nothing in the core code uses this, but we make it available to
15217 extensions (using mg_dup). */
15218 param->proto_perl = proto_perl;
15219 /* Likely nothing will use this, but it is initialised to be consistent
15220 with Perl_clone_params_new(). */
15221 param->new_perl = my_perl;
15222 param->unreferenced = NULL;
15225 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15227 PL_body_arenas = NULL;
15228 Zero(&PL_body_roots, 1, PL_body_roots);
15232 PL_sv_arenaroot = NULL;
15234 PL_debug = proto_perl->Idebug;
15236 /* dbargs array probably holds garbage */
15239 PL_compiling = proto_perl->Icompiling;
15241 /* pseudo environmental stuff */
15242 PL_origargc = proto_perl->Iorigargc;
15243 PL_origargv = proto_perl->Iorigargv;
15245 #ifndef NO_TAINT_SUPPORT
15246 /* Set tainting stuff before PerlIO_debug can possibly get called */
15247 PL_tainting = proto_perl->Itainting;
15248 PL_taint_warn = proto_perl->Itaint_warn;
15250 PL_tainting = FALSE;
15251 PL_taint_warn = FALSE;
15254 PL_minus_c = proto_perl->Iminus_c;
15256 PL_localpatches = proto_perl->Ilocalpatches;
15257 PL_splitstr = proto_perl->Isplitstr;
15258 PL_minus_n = proto_perl->Iminus_n;
15259 PL_minus_p = proto_perl->Iminus_p;
15260 PL_minus_l = proto_perl->Iminus_l;
15261 PL_minus_a = proto_perl->Iminus_a;
15262 PL_minus_E = proto_perl->Iminus_E;
15263 PL_minus_F = proto_perl->Iminus_F;
15264 PL_doswitches = proto_perl->Idoswitches;
15265 PL_dowarn = proto_perl->Idowarn;
15266 #ifdef PERL_SAWAMPERSAND
15267 PL_sawampersand = proto_perl->Isawampersand;
15269 PL_unsafe = proto_perl->Iunsafe;
15270 PL_perldb = proto_perl->Iperldb;
15271 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15272 PL_exit_flags = proto_perl->Iexit_flags;
15274 /* XXX time(&PL_basetime) when asked for? */
15275 PL_basetime = proto_perl->Ibasetime;
15277 PL_maxsysfd = proto_perl->Imaxsysfd;
15278 PL_statusvalue = proto_perl->Istatusvalue;
15280 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15282 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15285 /* RE engine related */
15286 PL_regmatch_slab = NULL;
15287 PL_reg_curpm = NULL;
15289 PL_sub_generation = proto_perl->Isub_generation;
15291 /* funky return mechanisms */
15292 PL_forkprocess = proto_perl->Iforkprocess;
15294 /* internal state */
15295 PL_main_start = proto_perl->Imain_start;
15296 PL_eval_root = proto_perl->Ieval_root;
15297 PL_eval_start = proto_perl->Ieval_start;
15299 PL_filemode = proto_perl->Ifilemode;
15300 PL_lastfd = proto_perl->Ilastfd;
15301 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15302 PL_gensym = proto_perl->Igensym;
15304 PL_laststatval = proto_perl->Ilaststatval;
15305 PL_laststype = proto_perl->Ilaststype;
15308 PL_profiledata = NULL;
15310 PL_generation = proto_perl->Igeneration;
15312 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15313 PL_in_clean_all = proto_perl->Iin_clean_all;
15315 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15316 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15317 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15318 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15319 PL_nomemok = proto_perl->Inomemok;
15320 PL_an = proto_perl->Ian;
15321 PL_evalseq = proto_perl->Ievalseq;
15322 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15323 PL_origalen = proto_perl->Iorigalen;
15325 PL_sighandlerp = proto_perl->Isighandlerp;
15327 PL_runops = proto_perl->Irunops;
15329 PL_subline = proto_perl->Isubline;
15331 PL_cv_has_eval = proto_perl->Icv_has_eval;
15334 PL_cryptseen = proto_perl->Icryptseen;
15337 #ifdef USE_LOCALE_COLLATE
15338 PL_collation_ix = proto_perl->Icollation_ix;
15339 PL_collation_standard = proto_perl->Icollation_standard;
15340 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15341 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15342 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15343 #endif /* USE_LOCALE_COLLATE */
15345 #ifdef USE_LOCALE_NUMERIC
15346 PL_numeric_standard = proto_perl->Inumeric_standard;
15347 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15348 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15349 #endif /* !USE_LOCALE_NUMERIC */
15351 /* Did the locale setup indicate UTF-8? */
15352 PL_utf8locale = proto_perl->Iutf8locale;
15353 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15354 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15355 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15356 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15357 PL_lc_numeric_mutex_depth = 0;
15359 /* Unicode features (see perlrun/-C) */
15360 PL_unicode = proto_perl->Iunicode;
15362 /* Pre-5.8 signals control */
15363 PL_signals = proto_perl->Isignals;
15365 /* times() ticks per second */
15366 PL_clocktick = proto_perl->Iclocktick;
15368 /* Recursion stopper for PerlIO_find_layer */
15369 PL_in_load_module = proto_perl->Iin_load_module;
15371 /* sort() routine */
15372 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15374 /* Not really needed/useful since the reenrant_retint is "volatile",
15375 * but do it for consistency's sake. */
15376 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15378 /* Hooks to shared SVs and locks. */
15379 PL_sharehook = proto_perl->Isharehook;
15380 PL_lockhook = proto_perl->Ilockhook;
15381 PL_unlockhook = proto_perl->Iunlockhook;
15382 PL_threadhook = proto_perl->Ithreadhook;
15383 PL_destroyhook = proto_perl->Idestroyhook;
15384 PL_signalhook = proto_perl->Isignalhook;
15386 PL_globhook = proto_perl->Iglobhook;
15389 PL_last_swash_hv = NULL; /* reinits on demand */
15390 PL_last_swash_klen = 0;
15391 PL_last_swash_key[0]= '\0';
15392 PL_last_swash_tmps = (U8*)NULL;
15393 PL_last_swash_slen = 0;
15395 PL_srand_called = proto_perl->Isrand_called;
15396 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15398 if (flags & CLONEf_COPY_STACKS) {
15399 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15400 PL_tmps_ix = proto_perl->Itmps_ix;
15401 PL_tmps_max = proto_perl->Itmps_max;
15402 PL_tmps_floor = proto_perl->Itmps_floor;
15404 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15405 * NOTE: unlike the others! */
15406 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15407 PL_scopestack_max = proto_perl->Iscopestack_max;
15409 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15410 * NOTE: unlike the others! */
15411 PL_savestack_ix = proto_perl->Isavestack_ix;
15412 PL_savestack_max = proto_perl->Isavestack_max;
15415 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15416 PL_top_env = &PL_start_env;
15418 PL_op = proto_perl->Iop;
15421 PL_Xpv = (XPV*)NULL;
15422 my_perl->Ina = proto_perl->Ina;
15424 PL_statcache = proto_perl->Istatcache;
15426 #ifndef NO_TAINT_SUPPORT
15427 PL_tainted = proto_perl->Itainted;
15429 PL_tainted = FALSE;
15431 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15433 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15435 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15436 PL_restartop = proto_perl->Irestartop;
15437 PL_in_eval = proto_perl->Iin_eval;
15438 PL_delaymagic = proto_perl->Idelaymagic;
15439 PL_phase = proto_perl->Iphase;
15440 PL_localizing = proto_perl->Ilocalizing;
15442 PL_hv_fetch_ent_mh = NULL;
15443 PL_modcount = proto_perl->Imodcount;
15444 PL_lastgotoprobe = NULL;
15445 PL_dumpindent = proto_perl->Idumpindent;
15447 PL_efloatbuf = NULL; /* reinits on demand */
15448 PL_efloatsize = 0; /* reinits on demand */
15452 PL_colorset = 0; /* reinits PL_colors[] */
15453 /*PL_colors[6] = {0,0,0,0,0,0};*/
15455 /* Pluggable optimizer */
15456 PL_peepp = proto_perl->Ipeepp;
15457 PL_rpeepp = proto_perl->Irpeepp;
15458 /* op_free() hook */
15459 PL_opfreehook = proto_perl->Iopfreehook;
15461 #ifdef USE_REENTRANT_API
15462 /* XXX: things like -Dm will segfault here in perlio, but doing
15463 * PERL_SET_CONTEXT(proto_perl);
15464 * breaks too many other things
15466 Perl_reentrant_init(aTHX);
15469 /* create SV map for pointer relocation */
15470 PL_ptr_table = ptr_table_new();
15472 /* initialize these special pointers as early as possible */
15474 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15475 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15476 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15477 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15478 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15479 &PL_padname_const);
15481 /* create (a non-shared!) shared string table */
15482 PL_strtab = newHV();
15483 HvSHAREKEYS_off(PL_strtab);
15484 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15485 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15487 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15489 /* This PV will be free'd special way so must set it same way op.c does */
15490 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15491 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15493 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15494 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15495 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15496 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15498 param->stashes = newAV(); /* Setup array of objects to call clone on */
15499 /* This makes no difference to the implementation, as it always pushes
15500 and shifts pointers to other SVs without changing their reference
15501 count, with the array becoming empty before it is freed. However, it
15502 makes it conceptually clear what is going on, and will avoid some
15503 work inside av.c, filling slots between AvFILL() and AvMAX() with
15504 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15505 AvREAL_off(param->stashes);
15507 if (!(flags & CLONEf_COPY_STACKS)) {
15508 param->unreferenced = newAV();
15511 #ifdef PERLIO_LAYERS
15512 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15513 PerlIO_clone(aTHX_ proto_perl, param);
15516 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15517 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15518 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15519 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15520 PL_xsubfilename = proto_perl->Ixsubfilename;
15521 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15522 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15525 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15526 PL_inplace = SAVEPV(proto_perl->Iinplace);
15527 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15529 /* magical thingies */
15531 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15532 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15533 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15536 /* Clone the regex array */
15537 /* ORANGE FIXME for plugins, probably in the SV dup code.
15538 newSViv(PTR2IV(CALLREGDUPE(
15539 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15541 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15542 PL_regex_pad = AvARRAY(PL_regex_padav);
15544 PL_stashpadmax = proto_perl->Istashpadmax;
15545 PL_stashpadix = proto_perl->Istashpadix ;
15546 Newx(PL_stashpad, PL_stashpadmax, HV *);
15549 for (; o < PL_stashpadmax; ++o)
15550 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15553 /* shortcuts to various I/O objects */
15554 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15555 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15556 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15557 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15558 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15559 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15560 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15562 /* shortcuts to regexp stuff */
15563 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15565 /* shortcuts to misc objects */
15566 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15568 /* shortcuts to debugging objects */
15569 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15570 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15571 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15572 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15573 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15574 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15575 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15577 /* symbol tables */
15578 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15579 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15580 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15581 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15582 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15584 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15585 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15586 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15587 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15588 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15589 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15590 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15591 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15592 PL_savebegin = proto_perl->Isavebegin;
15594 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15596 /* subprocess state */
15597 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15599 if (proto_perl->Iop_mask)
15600 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15603 /* PL_asserting = proto_perl->Iasserting; */
15605 /* current interpreter roots */
15606 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15608 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15611 /* runtime control stuff */
15612 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15614 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15616 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15618 /* interpreter atexit processing */
15619 PL_exitlistlen = proto_perl->Iexitlistlen;
15620 if (PL_exitlistlen) {
15621 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15622 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15625 PL_exitlist = (PerlExitListEntry*)NULL;
15627 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15628 if (PL_my_cxt_size) {
15629 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15630 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15633 PL_my_cxt_list = (void**)NULL;
15635 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15636 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15637 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15638 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15640 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15642 PAD_CLONE_VARS(proto_perl, param);
15644 #ifdef HAVE_INTERP_INTERN
15645 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15648 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15650 #ifdef PERL_USES_PL_PIDSTATUS
15651 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15653 PL_osname = SAVEPV(proto_perl->Iosname);
15654 PL_parser = parser_dup(proto_perl->Iparser, param);
15656 /* XXX this only works if the saved cop has already been cloned */
15657 if (proto_perl->Iparser) {
15658 PL_parser->saved_curcop = (COP*)any_dup(
15659 proto_perl->Iparser->saved_curcop,
15663 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15665 #if defined(USE_POSIX_2008_LOCALE) \
15666 && defined(USE_THREAD_SAFE_LOCALE) \
15667 && ! defined(HAS_QUERYLOCALE)
15668 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15669 PL_curlocales[i] = savepv("."); /* An illegal value */
15672 #ifdef USE_LOCALE_CTYPE
15673 /* Should we warn if uses locale? */
15674 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15677 #ifdef USE_LOCALE_COLLATE
15678 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15679 #endif /* USE_LOCALE_COLLATE */
15681 #ifdef USE_LOCALE_NUMERIC
15682 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15683 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15685 # if defined(HAS_POSIX_2008_LOCALE)
15686 PL_underlying_numeric_obj = NULL;
15688 #endif /* !USE_LOCALE_NUMERIC */
15690 PL_langinfo_buf = NULL;
15691 PL_langinfo_bufsize = 0;
15693 PL_setlocale_buf = NULL;
15694 PL_setlocale_bufsize = 0;
15696 /* utf8 character class swashes */
15697 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15699 if (proto_perl->Ipsig_pend) {
15700 Newxz(PL_psig_pend, SIG_SIZE, int);
15703 PL_psig_pend = (int*)NULL;
15706 if (proto_perl->Ipsig_name) {
15707 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15708 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15710 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15713 PL_psig_ptr = (SV**)NULL;
15714 PL_psig_name = (SV**)NULL;
15717 if (flags & CLONEf_COPY_STACKS) {
15718 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15719 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15720 PL_tmps_ix+1, param);
15722 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15723 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15724 Newx(PL_markstack, i, I32);
15725 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15726 - proto_perl->Imarkstack);
15727 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15728 - proto_perl->Imarkstack);
15729 Copy(proto_perl->Imarkstack, PL_markstack,
15730 PL_markstack_ptr - PL_markstack + 1, I32);
15732 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15733 * NOTE: unlike the others! */
15734 Newx(PL_scopestack, PL_scopestack_max, I32);
15735 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15738 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15739 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15741 /* reset stack AV to correct length before its duped via
15742 * PL_curstackinfo */
15743 AvFILLp(proto_perl->Icurstack) =
15744 proto_perl->Istack_sp - proto_perl->Istack_base;
15746 /* NOTE: si_dup() looks at PL_markstack */
15747 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15749 /* PL_curstack = PL_curstackinfo->si_stack; */
15750 PL_curstack = av_dup(proto_perl->Icurstack, param);
15751 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15753 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15754 PL_stack_base = AvARRAY(PL_curstack);
15755 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15756 - proto_perl->Istack_base);
15757 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15759 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15760 PL_savestack = ss_dup(proto_perl, param);
15764 ENTER; /* perl_destruct() wants to LEAVE; */
15767 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15768 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15770 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15771 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15772 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15773 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15774 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15775 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15777 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15779 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15780 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15781 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15783 PL_stashcache = newHV();
15785 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15786 proto_perl->Iwatchaddr);
15787 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15788 if (PL_debug && PL_watchaddr) {
15789 PerlIO_printf(Perl_debug_log,
15790 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15791 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15792 PTR2UV(PL_watchok));
15795 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15796 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15798 /* Call the ->CLONE method, if it exists, for each of the stashes
15799 identified by sv_dup() above.
15801 while(av_tindex(param->stashes) != -1) {
15802 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15803 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15804 if (cloner && GvCV(cloner)) {
15809 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15811 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15817 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15818 ptr_table_free(PL_ptr_table);
15819 PL_ptr_table = NULL;
15822 if (!(flags & CLONEf_COPY_STACKS)) {
15823 unreferenced_to_tmp_stack(param->unreferenced);
15826 SvREFCNT_dec(param->stashes);
15828 /* orphaned? eg threads->new inside BEGIN or use */
15829 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15830 SvREFCNT_inc_simple_void(PL_compcv);
15831 SAVEFREESV(PL_compcv);
15838 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15840 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15842 if (AvFILLp(unreferenced) > -1) {
15843 SV **svp = AvARRAY(unreferenced);
15844 SV **const last = svp + AvFILLp(unreferenced);
15848 if (SvREFCNT(*svp) == 1)
15850 } while (++svp <= last);
15852 EXTEND_MORTAL(count);
15853 svp = AvARRAY(unreferenced);
15856 if (SvREFCNT(*svp) == 1) {
15857 /* Our reference is the only one to this SV. This means that
15858 in this thread, the scalar effectively has a 0 reference.
15859 That doesn't work (cleanup never happens), so donate our
15860 reference to it onto the save stack. */
15861 PL_tmps_stack[++PL_tmps_ix] = *svp;
15863 /* As an optimisation, because we are already walking the
15864 entire array, instead of above doing either
15865 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15866 release our reference to the scalar, so that at the end of
15867 the array owns zero references to the scalars it happens to
15868 point to. We are effectively converting the array from
15869 AvREAL() on to AvREAL() off. This saves the av_clear()
15870 (triggered by the SvREFCNT_dec(unreferenced) below) from
15871 walking the array a second time. */
15872 SvREFCNT_dec(*svp);
15875 } while (++svp <= last);
15876 AvREAL_off(unreferenced);
15878 SvREFCNT_dec_NN(unreferenced);
15882 Perl_clone_params_del(CLONE_PARAMS *param)
15884 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15886 PerlInterpreter *const to = param->new_perl;
15888 PerlInterpreter *const was = PERL_GET_THX;
15890 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15896 SvREFCNT_dec(param->stashes);
15897 if (param->unreferenced)
15898 unreferenced_to_tmp_stack(param->unreferenced);
15908 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15911 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15912 does a dTHX; to get the context from thread local storage.
15913 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15914 a version that passes in my_perl. */
15915 PerlInterpreter *const was = PERL_GET_THX;
15916 CLONE_PARAMS *param;
15918 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15924 /* Given that we've set the context, we can do this unshared. */
15925 Newx(param, 1, CLONE_PARAMS);
15928 param->proto_perl = from;
15929 param->new_perl = to;
15930 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15931 AvREAL_off(param->stashes);
15932 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15940 #endif /* USE_ITHREADS */
15943 Perl_init_constants(pTHX)
15947 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15948 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15949 SvANY(&PL_sv_undef) = NULL;
15951 SvANY(&PL_sv_no) = new_XPVNV();
15952 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15953 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15954 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15957 SvANY(&PL_sv_yes) = new_XPVNV();
15958 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15959 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15960 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15963 SvANY(&PL_sv_zero) = new_XPVNV();
15964 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15965 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15966 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15970 SvPV_set(&PL_sv_no, (char*)PL_No);
15971 SvCUR_set(&PL_sv_no, 0);
15972 SvLEN_set(&PL_sv_no, 0);
15973 SvIV_set(&PL_sv_no, 0);
15974 SvNV_set(&PL_sv_no, 0);
15976 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15977 SvCUR_set(&PL_sv_yes, 1);
15978 SvLEN_set(&PL_sv_yes, 0);
15979 SvIV_set(&PL_sv_yes, 1);
15980 SvNV_set(&PL_sv_yes, 1);
15982 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15983 SvCUR_set(&PL_sv_zero, 1);
15984 SvLEN_set(&PL_sv_zero, 0);
15985 SvIV_set(&PL_sv_zero, 0);
15986 SvNV_set(&PL_sv_zero, 0);
15988 PadnamePV(&PL_padname_const) = (char *)PL_No;
15990 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15991 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15992 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15993 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15995 assert(SvIMMORTAL(&PL_sv_yes));
15996 assert(SvIMMORTAL(&PL_sv_undef));
15997 assert(SvIMMORTAL(&PL_sv_no));
15998 assert(SvIMMORTAL(&PL_sv_zero));
16000 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
16001 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
16002 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
16003 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
16005 assert( SvTRUE_nomg_NN(&PL_sv_yes));
16006 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
16007 assert(!SvTRUE_nomg_NN(&PL_sv_no));
16008 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
16012 =head1 Unicode Support
16014 =for apidoc sv_recode_to_utf8
16016 C<encoding> is assumed to be an C<Encode> object, on entry the PV
16017 of C<sv> is assumed to be octets in that encoding, and C<sv>
16018 will be converted into Unicode (and UTF-8).
16020 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
16021 is not a reference, nothing is done to C<sv>. If C<encoding> is not
16022 an C<Encode::XS> Encoding object, bad things will happen.
16023 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
16025 The PV of C<sv> is returned.
16030 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
16032 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
16034 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
16043 if (SvPADTMP(nsv)) {
16044 nsv = sv_newmortal();
16045 SvSetSV_nosteal(nsv, sv);
16054 Passing sv_yes is wrong - it needs to be or'ed set of constants
16055 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16056 remove converted chars from source.
16058 Both will default the value - let them.
16060 XPUSHs(&PL_sv_yes);
16063 call_method("decode", G_SCALAR);
16067 s = SvPV_const(uni, len);
16068 if (s != SvPVX_const(sv)) {
16069 SvGROW(sv, len + 1);
16070 Move(s, SvPVX(sv), len + 1, char);
16071 SvCUR_set(sv, len);
16076 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16077 /* clear pos and any utf8 cache */
16078 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16081 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16082 magic_setutf8(sv,mg); /* clear UTF8 cache */
16087 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16091 =for apidoc sv_cat_decode
16093 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16094 assumed to be octets in that encoding and decoding the input starts
16095 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16096 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16097 when the string C<tstr> appears in decoding output or the input ends on
16098 the PV of C<ssv>. The value which C<offset> points will be modified
16099 to the last input position on C<ssv>.
16101 Returns TRUE if the terminator was found, else returns FALSE.
16106 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16107 SV *ssv, int *offset, char *tstr, int tlen)
16111 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16113 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16124 offsv = newSViv(*offset);
16126 mPUSHp(tstr, tlen);
16128 call_method("cat_decode", G_SCALAR);
16130 ret = SvTRUE(TOPs);
16131 *offset = SvIV(offsv);
16137 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16142 /* ---------------------------------------------------------------------
16144 * support functions for report_uninit()
16147 /* the maxiumum size of array or hash where we will scan looking
16148 * for the undefined element that triggered the warning */
16150 #define FUV_MAX_SEARCH_SIZE 1000
16152 /* Look for an entry in the hash whose value has the same SV as val;
16153 * If so, return a mortal copy of the key. */
16156 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16162 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16164 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16165 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16168 array = HvARRAY(hv);
16170 for (i=HvMAX(hv); i>=0; i--) {
16172 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16173 if (HeVAL(entry) != val)
16175 if ( HeVAL(entry) == &PL_sv_undef ||
16176 HeVAL(entry) == &PL_sv_placeholder)
16180 if (HeKLEN(entry) == HEf_SVKEY)
16181 return sv_mortalcopy(HeKEY_sv(entry));
16182 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16188 /* Look for an entry in the array whose value has the same SV as val;
16189 * If so, return the index, otherwise return -1. */
16192 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16194 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16196 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16197 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16200 if (val != &PL_sv_undef) {
16201 SV ** const svp = AvARRAY(av);
16204 for (i=AvFILLp(av); i>=0; i--)
16211 /* varname(): return the name of a variable, optionally with a subscript.
16212 * If gv is non-zero, use the name of that global, along with gvtype (one
16213 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16214 * targ. Depending on the value of the subscript_type flag, return:
16217 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16218 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16219 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16220 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16223 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16224 const SV *const keyname, SSize_t aindex, int subscript_type)
16227 SV * const name = sv_newmortal();
16228 if (gv && isGV(gv)) {
16230 buffer[0] = gvtype;
16233 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16235 gv_fullname4(name, gv, buffer, 0);
16237 if ((unsigned int)SvPVX(name)[1] <= 26) {
16239 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16241 /* Swap the 1 unprintable control character for the 2 byte pretty
16242 version - ie substr($name, 1, 1) = $buffer; */
16243 sv_insert(name, 1, 1, buffer, 2);
16247 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16250 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16252 if (!cv || !CvPADLIST(cv))
16254 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16255 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16259 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16260 SV * const sv = newSV(0);
16262 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16264 *SvPVX(name) = '$';
16265 Perl_sv_catpvf(aTHX_ name, "{%s}",
16266 pv_pretty(sv, pv, len, 32, NULL, NULL,
16267 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16268 SvREFCNT_dec_NN(sv);
16270 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16271 *SvPVX(name) = '$';
16272 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16274 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16275 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16276 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16284 =for apidoc find_uninit_var
16286 Find the name of the undefined variable (if any) that caused the operator
16287 to issue a "Use of uninitialized value" warning.
16288 If match is true, only return a name if its value matches C<uninit_sv>.
16289 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16290 warning, then following the direct child of the op may yield an
16291 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16292 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16293 the variable name if we get an exact match.
16294 C<desc_p> points to a string pointer holding the description of the op.
16295 This may be updated if needed.
16297 The name is returned as a mortal SV.
16299 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16300 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16306 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16307 bool match, const char **desc_p)
16312 const OP *o, *o2, *kid;
16314 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16316 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16317 uninit_sv == &PL_sv_placeholder)))
16320 switch (obase->op_type) {
16323 /* undef should care if its args are undef - any warnings
16324 * will be from tied/magic vars */
16332 const bool pad = ( obase->op_type == OP_PADAV
16333 || obase->op_type == OP_PADHV
16334 || obase->op_type == OP_PADRANGE
16337 const bool hash = ( obase->op_type == OP_PADHV
16338 || obase->op_type == OP_RV2HV
16339 || (obase->op_type == OP_PADRANGE
16340 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16344 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16346 if (pad) { /* @lex, %lex */
16347 sv = PAD_SVl(obase->op_targ);
16351 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16352 /* @global, %global */
16353 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16356 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16358 else if (obase == PL_op) /* @{expr}, %{expr} */
16359 return find_uninit_var(cUNOPx(obase)->op_first,
16360 uninit_sv, match, desc_p);
16361 else /* @{expr}, %{expr} as a sub-expression */
16365 /* attempt to find a match within the aggregate */
16367 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16369 subscript_type = FUV_SUBSCRIPT_HASH;
16372 index = find_array_subscript((const AV *)sv, uninit_sv);
16374 subscript_type = FUV_SUBSCRIPT_ARRAY;
16377 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16380 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16381 keysv, index, subscript_type);
16385 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16387 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16388 if (!gv || !GvSTASH(gv))
16390 if (match && (GvSV(gv) != uninit_sv))
16392 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16395 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16398 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16400 return varname(NULL, '$', obase->op_targ,
16401 NULL, 0, FUV_SUBSCRIPT_NONE);
16404 gv = cGVOPx_gv(obase);
16405 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16407 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16409 case OP_AELEMFAST_LEX:
16412 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16413 if (!av || SvRMAGICAL(av))
16415 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16416 if (!svp || *svp != uninit_sv)
16419 return varname(NULL, '$', obase->op_targ,
16420 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16423 gv = cGVOPx_gv(obase);
16428 AV *const av = GvAV(gv);
16429 if (!av || SvRMAGICAL(av))
16431 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16432 if (!svp || *svp != uninit_sv)
16435 return varname(gv, '$', 0,
16436 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16438 NOT_REACHED; /* NOTREACHED */
16441 o = cUNOPx(obase)->op_first;
16442 if (!o || o->op_type != OP_NULL ||
16443 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16445 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16450 bool negate = FALSE;
16452 if (PL_op == obase)
16453 /* $a[uninit_expr] or $h{uninit_expr} */
16454 return find_uninit_var(cBINOPx(obase)->op_last,
16455 uninit_sv, match, desc_p);
16458 o = cBINOPx(obase)->op_first;
16459 kid = cBINOPx(obase)->op_last;
16461 /* get the av or hv, and optionally the gv */
16463 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16464 sv = PAD_SV(o->op_targ);
16466 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16467 && cUNOPo->op_first->op_type == OP_GV)
16469 gv = cGVOPx_gv(cUNOPo->op_first);
16473 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16478 if (kid && kid->op_type == OP_NEGATE) {
16480 kid = cUNOPx(kid)->op_first;
16483 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16484 /* index is constant */
16487 kidsv = newSVpvs_flags("-", SVs_TEMP);
16488 sv_catsv(kidsv, cSVOPx_sv(kid));
16491 kidsv = cSVOPx_sv(kid);
16495 if (obase->op_type == OP_HELEM) {
16496 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16497 if (!he || HeVAL(he) != uninit_sv)
16501 SV * const opsv = cSVOPx_sv(kid);
16502 const IV opsviv = SvIV(opsv);
16503 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16504 negate ? - opsviv : opsviv,
16506 if (!svp || *svp != uninit_sv)
16510 if (obase->op_type == OP_HELEM)
16511 return varname(gv, '%', o->op_targ,
16512 kidsv, 0, FUV_SUBSCRIPT_HASH);
16514 return varname(gv, '@', o->op_targ, NULL,
16515 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16516 FUV_SUBSCRIPT_ARRAY);
16519 /* index is an expression;
16520 * attempt to find a match within the aggregate */
16521 if (obase->op_type == OP_HELEM) {
16522 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16524 return varname(gv, '%', o->op_targ,
16525 keysv, 0, FUV_SUBSCRIPT_HASH);
16528 const SSize_t index
16529 = find_array_subscript((const AV *)sv, uninit_sv);
16531 return varname(gv, '@', o->op_targ,
16532 NULL, index, FUV_SUBSCRIPT_ARRAY);
16537 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16539 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16541 NOT_REACHED; /* NOTREACHED */
16544 case OP_MULTIDEREF: {
16545 /* If we were executing OP_MULTIDEREF when the undef warning
16546 * triggered, then it must be one of the index values within
16547 * that triggered it. If not, then the only possibility is that
16548 * the value retrieved by the last aggregate index might be the
16549 * culprit. For the former, we set PL_multideref_pc each time before
16550 * using an index, so work though the item list until we reach
16551 * that point. For the latter, just work through the entire item
16552 * list; the last aggregate retrieved will be the candidate.
16553 * There is a third rare possibility: something triggered
16554 * magic while fetching an array/hash element. Just display
16555 * nothing in this case.
16558 /* the named aggregate, if any */
16559 PADOFFSET agg_targ = 0;
16561 /* the last-seen index */
16563 PADOFFSET index_targ;
16565 IV index_const_iv = 0; /* init for spurious compiler warn */
16566 SV *index_const_sv;
16567 int depth = 0; /* how many array/hash lookups we've done */
16569 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16570 UNOP_AUX_item *last = NULL;
16571 UV actions = items->uv;
16574 if (PL_op == obase) {
16575 last = PL_multideref_pc;
16576 assert(last >= items && last <= items + items[-1].uv);
16583 switch (actions & MDEREF_ACTION_MASK) {
16585 case MDEREF_reload:
16586 actions = (++items)->uv;
16589 case MDEREF_HV_padhv_helem: /* $lex{...} */
16592 case MDEREF_AV_padav_aelem: /* $lex[...] */
16593 agg_targ = (++items)->pad_offset;
16597 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16600 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16602 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16603 assert(isGV_with_GP(agg_gv));
16606 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16607 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16610 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16611 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16617 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16618 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16621 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16622 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16629 index_const_sv = NULL;
16631 index_type = (actions & MDEREF_INDEX_MASK);
16632 switch (index_type) {
16633 case MDEREF_INDEX_none:
16635 case MDEREF_INDEX_const:
16637 index_const_sv = UNOP_AUX_item_sv(++items)
16639 index_const_iv = (++items)->iv;
16641 case MDEREF_INDEX_padsv:
16642 index_targ = (++items)->pad_offset;
16644 case MDEREF_INDEX_gvsv:
16645 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16646 assert(isGV_with_GP(index_gv));
16650 if (index_type != MDEREF_INDEX_none)
16653 if ( index_type == MDEREF_INDEX_none
16654 || (actions & MDEREF_FLAG_last)
16655 || (last && items >= last)
16659 actions >>= MDEREF_SHIFT;
16662 if (PL_op == obase) {
16663 /* most likely index was undef */
16665 *desc_p = ( (actions & MDEREF_FLAG_last)
16666 && (obase->op_private
16667 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16669 (obase->op_private & OPpMULTIDEREF_EXISTS)
16672 : is_hv ? "hash element" : "array element";
16673 assert(index_type != MDEREF_INDEX_none);
16675 if (GvSV(index_gv) == uninit_sv)
16676 return varname(index_gv, '$', 0, NULL, 0,
16677 FUV_SUBSCRIPT_NONE);
16682 if (PL_curpad[index_targ] == uninit_sv)
16683 return varname(NULL, '$', index_targ,
16684 NULL, 0, FUV_SUBSCRIPT_NONE);
16688 /* If we got to this point it was undef on a const subscript,
16689 * so magic probably involved, e.g. $ISA[0]. Give up. */
16693 /* the SV returned by pp_multideref() was undef, if anything was */
16699 sv = PAD_SV(agg_targ);
16701 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16708 if (index_type == MDEREF_INDEX_const) {
16713 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16714 if (!he || HeVAL(he) != uninit_sv)
16718 SV * const * const svp =
16719 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16720 if (!svp || *svp != uninit_sv)
16725 ? varname(agg_gv, '%', agg_targ,
16726 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16727 : varname(agg_gv, '@', agg_targ,
16728 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16731 /* index is an var */
16733 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16735 return varname(agg_gv, '%', agg_targ,
16736 keysv, 0, FUV_SUBSCRIPT_HASH);
16739 const SSize_t index
16740 = find_array_subscript((const AV *)sv, uninit_sv);
16742 return varname(agg_gv, '@', agg_targ,
16743 NULL, index, FUV_SUBSCRIPT_ARRAY);
16747 return varname(agg_gv,
16749 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16751 NOT_REACHED; /* NOTREACHED */
16755 /* only examine RHS */
16756 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16760 o = cUNOPx(obase)->op_first;
16761 if ( o->op_type == OP_PUSHMARK
16762 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16766 if (!OpHAS_SIBLING(o)) {
16767 /* one-arg version of open is highly magical */
16769 if (o->op_type == OP_GV) { /* open FOO; */
16771 if (match && GvSV(gv) != uninit_sv)
16773 return varname(gv, '$', 0,
16774 NULL, 0, FUV_SUBSCRIPT_NONE);
16776 /* other possibilities not handled are:
16777 * open $x; or open my $x; should return '${*$x}'
16778 * open expr; should return '$'.expr ideally
16785 /* ops where $_ may be an implicit arg */
16790 if ( !(obase->op_flags & OPf_STACKED)) {
16791 if (uninit_sv == DEFSV)
16792 return newSVpvs_flags("$_", SVs_TEMP);
16793 else if (obase->op_targ
16794 && uninit_sv == PAD_SVl(obase->op_targ))
16795 return varname(NULL, '$', obase->op_targ, NULL, 0,
16796 FUV_SUBSCRIPT_NONE);
16803 match = 1; /* print etc can return undef on defined args */
16804 /* skip filehandle as it can't produce 'undef' warning */
16805 o = cUNOPx(obase)->op_first;
16806 if ((obase->op_flags & OPf_STACKED)
16808 ( o->op_type == OP_PUSHMARK
16809 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16810 o = OpSIBLING(OpSIBLING(o));
16814 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16815 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16817 /* the following ops are capable of returning PL_sv_undef even for
16818 * defined arg(s) */
16837 case OP_GETPEERNAME:
16884 case OP_SMARTMATCH:
16893 /* XXX tmp hack: these two may call an XS sub, and currently
16894 XS subs don't have a SUB entry on the context stack, so CV and
16895 pad determination goes wrong, and BAD things happen. So, just
16896 don't try to determine the value under those circumstances.
16897 Need a better fix at dome point. DAPM 11/2007 */
16903 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16904 if (gv && GvSV(gv) == uninit_sv)
16905 return newSVpvs_flags("$.", SVs_TEMP);
16910 /* def-ness of rval pos() is independent of the def-ness of its arg */
16911 if ( !(obase->op_flags & OPf_MOD))
16917 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16918 return newSVpvs_flags("${$/}", SVs_TEMP);
16923 if (!(obase->op_flags & OPf_KIDS))
16925 o = cUNOPx(obase)->op_first;
16931 /* This loop checks all the kid ops, skipping any that cannot pos-
16932 * sibly be responsible for the uninitialized value; i.e., defined
16933 * constants and ops that return nothing. If there is only one op
16934 * left that is not skipped, then we *know* it is responsible for
16935 * the uninitialized value. If there is more than one op left, we
16936 * have to look for an exact match in the while() loop below.
16937 * Note that we skip padrange, because the individual pad ops that
16938 * it replaced are still in the tree, so we work on them instead.
16941 for (kid=o; kid; kid = OpSIBLING(kid)) {
16942 const OPCODE type = kid->op_type;
16943 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16944 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16945 || (type == OP_PUSHMARK)
16946 || (type == OP_PADRANGE)
16950 if (o2) { /* more than one found */
16957 return find_uninit_var(o2, uninit_sv, match, desc_p);
16959 /* scan all args */
16961 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16973 =for apidoc report_uninit
16975 Print appropriate "Use of uninitialized variable" warning.
16981 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16983 const char *desc = NULL;
16984 SV* varname = NULL;
16987 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16989 : PL_op->op_type == OP_MULTICONCAT
16990 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16993 if (uninit_sv && PL_curpad) {
16994 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16996 sv_insert(varname, 0, 0, " ", 1);
16999 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
17000 /* we've reached the end of a sort block or sub,
17001 * and the uninit value is probably what that code returned */
17004 /* PL_warn_uninit_sv is constant */
17005 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
17007 /* diag_listed_as: Use of uninitialized value%s */
17008 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
17009 SVfARG(varname ? varname : &PL_sv_no),
17012 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
17014 GCC_DIAG_RESTORE_STMT;
17018 * ex: set ts=8 sts=4 sw=4 et: